The past two weeks have been nothing short of extraordinary as we've witnessed the successful implementation of Hachiko across veterinary teams. The enthusiasm and rapid adoption have far exceeded our expectations, demonstrating just how much the industry needed a solution like ours.

From day one, we've seen veterinary professionals embrace Hachiko's capabilities, integrating it seamlessly into their daily workflows. The feedback has been overwhelmingly positive, with teams reporting significant improvements in their operational efficiency and patient care delivery.

How Clinical Teams Use Hachiko:

• Pre-review of clinical records in GP or referral settings

• Quick summaries for GPs to submit referrals

• Rapid assessment of a clinical record in emergency settings

• Improved case transfer between clinicians and clinical record review

• Enabling faster diagnostic & treatment requests within the hospital

• Writing faster discharge summaries and getting home sooner

Saving Time

In just two weeks since our launch, ANI.ML Health has generated over 200 medical summaries, resulting in approximately 50 hours saved for veterinary teams.

This achievement represents more than just numbers - it's a testament to our commitment to streamlining care delivery processes and enabling medical professionals to focus more on patient care. Each hour saved is an hour that veterinary healthcare providers can dedicate to direct patient interaction, complex case analysis, or heading home sooner.

Did You Know?

You can submit feedback right within the app. Rate the summary out of 5 at the time of generation, or in your history tab in case you forgot. This gives us, and the model direct feedback so it can continually improve.

The Road Ahead

These initial results validate our mission and demonstrate the significant impact our technology can have on healthcare delivery. We're committed to building on this success and continuing to innovate in ways that make healthcare more efficient and effective for everyone involved.

We want to thank our early adopters and partners who have helped us achieve this milestone. Your trust and feedback have been invaluable in refining our service and delivering these impressive results.

Feature Requests?

We are actively working with all of our teams to continue to improve Hachiko every week. Our team is hard at work but we welcome any feedback and feature requests to help improve our development pipeline.

Email Us: we.care@animl.health 

Introducing Hachiko: Your AI-Powered Medical Record Summarizer for Veterinary Teams

In our fast-paced world of veterinary medicine, every second counts. Whether you're managing multiple cases in a busy practice, reviewing complex medical histories, or communicating with colleagues, the burden of documentation can be overwhelming. That's where Hachiko, our AI-powered medical record summarizer, comes in to revolutionize how veterinary teams interact with patient records.

WE ARE LIVE!

_{Click link above to sign up}

Why We Built Hachiko

As veterinary professionals, we understand the challenges you face. Sorting through lengthy medical records to extract relevant clinical insights is time-consuming and can take valuable time away from patient care or getting home on time. Our team at ANI.ML Health saw an opportunity to harness the power of artificial intelligence to streamline this process, enabling veterinary teams to spend less time reading and get home to your family sooner.

We are the first team to tackle this problem and the uniqueness of veterinary records. We are not just a ChatGPT wrapper and instead, our models are built by you and for you simply by using the summarizer.

What is Hachiko?

Hachiko is an advanced AI-driven platform designed to summarize complex veterinary medical records into clear, concise, and actionable insights. Whether you're reviewing a referral case or preparing for a client visit, Hachiko helps you quickly grasp the most critical details without the hassle of combing through pages of documentation.

Key Features

• Automated Summaries: Quickly generate concise summaries of medical records with a single click.

• Time-Saving Efficiency: Reduce the time spent reviewing lengthy records, allowing for more focus on patient care.

• Enhanced Collaboration: Improve communication between general practitioners, specialists, and support teams.

• Smart Insights: Extract relevant information such as diagnosis, treatment history, and current medications.

And so much more to come!

How Hachiko Works

Hachiko is simple to use and seamlessly integrates into your workflow:

1. Upload Medical Records: Simply upload patient records in PDF format.

2. AI-Powered Summarization: Our advanced veterinary-based algorithms analyze and condense the key information.

3. Review and Share: Access concise summaries and share them with your team for better decision-making.

Who Can Benefit from Hachiko?

• General Practitioners looking to get a quick overview of cases,

• Specialists reviewing complex medical histories and referrals,

• Veterinary Technicians ensuring all patient details are readily available,

• Resident, Intern, or Student preparing for any of your cases,

• Practice Manager streamlining workflows and improving efficiency

Hachiko is designed to support your team!

Join the Future of Veterinary Documentation

We’re excited to bring Hachiko to veterinary teams who are ready to embrace the future of AI-driven efficiency. Sign up now for early access and be among the first to experience how Hachiko can transform your practice.

🔗 Sign up now!

📧 we.care@animl.health

Stay tuned for more updates, insights, and success stories as we launch Hachiko into the veterinary world!

Parsing Veterinary Text With A Human Medical Model

We’re thrilled to share the findings from our latest research published in Veterinary and Comparative Oncology, titled: "Precision in Parsing: Evaluation of an Open-Source Named Entity Recognizer (NER) in Veterinary Oncology."

This study explores the exciting potential of artificial intelligence (AI) in veterinary medicine, specifically the application of Named Entity Recognition (NER) to analyze unstructured veterinary oncology medical records. Using the Bio-Epidemiology NER (BioEN), an open-source AI tool developed for human medical text, we evaluated its ability to identify and categorize entities within veterinary records, comparing its performance to human annotators.

Key Highlights:

• Why It Matters: Veterinary teams generate extensive clinical data, and tools like NER could streamline data extraction, reducing the manual workload for veterinary teams and opening new possibilities for summarization, clinical trial matching, treatment optimization, and more.

• Findings: While the BioEN tool demonstrated promise in parsing text quickly and categorizing entities, it faced challenges with veterinary-specific terminologies, such as surgical descriptors and bloodwork metrics, emphasizing the need for tailored veterinary AI solutions.

• Impact: This work highlights the potential of leveraging AI for more efficient clinical decision-making and the importance of developing veterinary-specific language tools.

This study is a step forward in bridging the gap between AI applications in human healthcare and veterinary medicine, fostering a future where data-driven decisions enhance patient outcomes in veterinary oncology.

To learn more, access the full study here or explore the code and resources on our GitHub repository.

What’s next for us?

Our recently published study on Named Entity Recognition (NER) in veterinary oncology demonstrated the potential of AI to transform how we handle unstructured medical records, but also highlighted the failure points of current AI tools in the market. But we're not stopping there.

We’re thrilled to introduce Hachiko, the first of its kind—an AI-powered medical record summarizer built specifically for veterinary medicine. Unlike current general AI solutions, Hachiko was developed with a veterinary-specific pipeline, designed to ensure unparalleled clinical utility and relevance for clinical teams.

Inspired by the loyalty and devotion of Hachiko, the famous Japanese dog who waited every day for his owner who never returned; we understand that long hours and demanding workflows often mean late nights at the clinic

This groundbreaking tool is set to revolutionize how veterinary teams manage medical records, allowing more time to focus on what matters most—patient care & getting home on time to take care of yourself.

Change is coming. 🐾

Sign up to hear more and prepare for launch in January 2025 🚀

Cygnus and Balto Partner To Bring You Discounted Specialty Services

At ANI.ML Health, our mission is to empower veterinary professionals with the tools and expertise they need to provide exceptional care to their patients. Today, we’re excited to announce a new partnership with Cygnus Veterinary Imaging Group, a leader in advanced diagnostic imaging reporting. Together, we’re taking another step to make specialty consultations more accessible, efficient, and affordable in Canada.

What Does This Partnership Mean for You?

As part of this collaboration, Cygnus clients will now receive 10% off Balto Specialty Consultations. This exclusive discount is available for all cases reviewed by Cygnus and is designed to streamline the referral process while providing cost-effective support for complex cases.

Our goal is to make it easier for you to connect with Balto’s team of board-certified specialists. Whether you’re seeking guidance on a challenging diagnosis or looking for input on treatment and anesthetic plans, Balto is here to support you and your patients.

How to Redeem the Discount

Taking advantage of this offer is simple:

1. Validate Your Discount: email we.care@animl.health with your Cygnus case number to verify your eligibility. You will receive an special discount code via email.

2. Visit Balto Specialty Consults: Head to https://consult.animl.health to request a consultation.

3. Submit Your Case Details: Include your Cygnus diagnostic report and case number when submitting your request.

Why This Matters

Veterinary medicine is a team effort, and partnerships like this allow us to better support clinicians on the front lines. Whether you're managing a patient with a complex condition or seeking specialized input to refine your treatment plan, our collaboration with Cygnus ensures you’ll have access to high-quality imaging insights combined with Balto’s specialty expertise.

This partnership represents our shared commitment to supporting veterinary professionals with the tools and expertise they need. Together with Cygnus, we’re making specialty consultations more accessible, affordable, and impactful.

Have questions or want to learn more? Reach out to us at we.care@animl.health, and let’s work together to elevate patient care!

ANI.ML Health Has Joined the NEXT AI Bootcamp: Driving Innovation In Veterinary AI

We’re so excited to announce that ANI.ML Health has been selected for the prestigious NEXTAI Bootcamp, a leading program supporting AI-driven startups. This milestone reflects our commitment to transforming veterinary healthcare through innovative AI solutions.

What is NEXTAI?

NEXTAI is a world-class accelerator empowering entrepreneurs to scale AI-based technologies. Through mentorship, resources, and industry connections, it helps startups bring impactful innovations to market.

Joining the NEXTAI bootcamp is a big step toward our mission of revolutionizing veterinary care. We’re eager to share updates as we develop new tools to empower veterinary professionals and improve patient outcomes. Stay tuned for more, and feel free to reach out to learn more about ANI.ML Health and our solutions!

Don’t forget our new additions!

We’re thrilled that our platform now offers remote consultations in dermatology and cardiology—and we’re just as excited as the day we launched them!

From tricky skin conditions to complex cardiac cases, our board-certified specialists are here to support you with expert guidance. Yes, that even includes remote echocardiogram evaluations!

These new services are designed to make specialty expertise more accessible than ever, so you can provide the best care for your patients without the need for travel or referrals.

🔗 Sign up today at consult.animl.health to take advantage of these exciting additions to our specialty consultation services.

Let us help you tackle those challenging cases in dermatology and cardiology—because every patient deserves top-tier care!

New Specialists & Specialties Added to Balto

After popular demand, we’ve added another specialty this week, Cardiology!

Dr. Sonja Fonfara joins us from the Ontario Veterinary College and will be providing consultation starting on November 4th, 2024. Head to Balto, sign up for a free account, and get access to veterinary specialists!

Dr. Sonja Fonfara DVM, Dr med vet, PhD, CertVC, DECVIM (Cardiology)

We also wanted to welcome Dr. Sabrina Ayoub, a specialist in critical care who has joined the Balto Specialty Consult team and will be available starting next week. Dr. Ayoub offers consultations in both English AND French for those of you in Quebec!

Dr. Sabrina Ayoub, DVM DACVECC

We have a few other exciting team updates coming in the next few weeks, stay tuned!

Have feedback or are looking for more support? Get in touch with us at:

we.care@animl.health

Anesthesia CE Event

We were thrilled to host our recent CE event, "Anesthesia After Hours," on October 30th, and the community response has been overwhelmingly positive. A huge thank you to Dr. Natalie Chow and Tasha Ford for delivering an engaging, case-based learning experience that resonated with our veterinary colleagues.

If you attended and have any feedback, or if you'd like to suggest topics or specialties for future CE sessions, please share your thoughts through the form below!

For those interested in a recording, we’re excited to announce that a recorded version of this session will be available for purchase starting next week. And, upon request, we’re happy to provide a CE certificate as well!

Fill Out Our Feedback Form Here

We’ve got some big news coming for our next newsletter, so stay tuned!

That’s it for now, but never hesitate to reach out to us with any feedback or if you want to learn more.

Chris

Anesthesia After Dark: Tricks, Treats, and Tips for Veterinary Teams – October 30th Event

Register Here for "Anesthesia After Dark" and secure your spot!

Are you ready for an evening filled with veterinary anesthesia insights? This Halloween season, we’re thrilled to invite you to a special event designed just for veterinary teams: "Anesthesia After Dark: Tricks, Treats, and Tips for Veterinary Teams."

This educational session, in collaboration with Scribenote, and Obi Veterinary Education, will take place on October 30th, 2024, from 6:30 PM to 8:00 PM.

What to Expect

The session will be led by two anesthesia experts who are passionate about helping veterinary teams master the art of anesthesia:

• Dr. Natalie Chow, a highly experienced anesthesiologist, will guide us through essential tips and techniques to ensure smooth and safe anesthesia management in clinical settings.

• Natasha Ford, RVT specializing in anesthesia, will provide practical insights from her experience in handling anesthesia across various cases. Her hands-on approach will give veterinary technicians the confidence to implement these strategies in their own practices.

Why Attend?

Anesthesia can sometimes feel like walking into the unknown, but our experts are here to show you that it doesn’t have to be frightening! With a focus on delivering tricks to avoid anesthesia-related challenges and treats in the form of easy-to-apply tips, this event promises to provide valuable knowledge that you can immediately bring back to your practice.

Whether you're a veterinarian, veterinary technician, or part of a vet care team, this evening will give you practical advice on managing anesthesia during those tricky cases that may come your way.

Registration Details

Don’t miss out on this exciting event that combines professional development with Halloween fun!

Date: October 30th, 2024
Time: 6:30 PM – 8:00 PM
Location: Virtual Event

Register Here for "Anesthesia After Dark" and secure your spot!

About Our Collaborators:

• Balto Specialty Consults: Balto provides 24/7 access to veterinary specialists, offering remote consultations for anesthesia, internal medicine, surgery, oncology, and more.

• Scribenote: A revolutionary AI scribe tool that makes veterinary documentation a breeze, allowing you to focus more on patient care.

• Obi Veterinary Education: Dedicated to advancing veterinary learning through high-quality educational content tailored to veterinary professionals.

Dermatology is now on Balto!

We want to welcome Dr. Neoklis Apostolopoulos as our first Dermatologist to the Balto team! Veterinary teams asked for it, and we delivered. Dermatology cases account for the second highest case presentations in general practice, only over wellness visits. This level of expertise helps us set Balto apart and we are so excited to have him with us!

He is available every week for remote dermatology consultation with veterinary teams and he’ll be working with our research team on many of our projects given his background and interest in computer vision.

A little about Dr. Apostolopoulos:

Dr. Neoklis Apostolopoulos obtained his veterinary degree from the University of Thessaly in Greece. He then relocated to Germany, where he completed a rotating internship, a residency, and postgraduate research focusing on canine cutaneous microbiota. Currently, he serves as an assistant professor of veterinary dermatology at the Louisiana State University – School of Veterinary Medicine. Dr. Apostolopoulos is the principal investigator of NAVDLab, where his research centers on applied artificial intelligence in veterinary dermatology and the canine skin microbiome in health and disease. His clinical interests encompass all aspects of dermatology, allergology, and otology. Dr. Apostolopoulos also has extensive experience in CO2 laser surgery and fluorescence photobiomodulation therapy for treating skin diseases.

A full list of his publications can be found here: https://orcid.org/0000-0001-9749-0795

Meet Tyler Poore, ANI.ML Health’s new Machine Learning Engineer

We wanted to take a moment to announce the newest member of the ANIML Health team, Tyler Poore!

Tyler brings a wealth of knowledge and experience in biological and medical AI and we couldn't be more excited to have him join us to help our team tackle some of the biggest problems in veterinary healthcare delivery.

A little about Tyler:

Tyler Poore is a Machine Learning Engineer specializing in natural language processing, biotechnology, and computer vision. With a foundation in chemistry from Columbia University, Tyler left his PhD program to co-found ExSponge, Inc., focusing on antimicrobial polymer synthesis. He then helped start IrOs, LLC, where he led R&D efforts in developing photo-catalyzed hydrogel polymers for surgical applications. It was during his time at the startup that he discovered his passion for data science and machine learning. His current work involves designing and deploying ML models with a focus on LLMs for practical applications in healthcare and finance. As an instructor at Inspirit AI, Tyler teaches advanced courses on deep learning, including U-nets, GANs, and Transformers. Originally from Montana, Tyler balances his tech-focused career with a love for outdoor activities, particularly hiking and downhill skiing.

We’re super excited to have our team grow and we have a few other announcements coming soon. Stay tuned!

ANI.ML Health has been accepted into the University of Toronto's Health to Innovation Hub Program!

We are thrilled to announce that ANI.ML Health has been accepted into the prestigious Health Innovation Hub (H2I) at the University of Toronto! This opportunity marks an exciting chapter for our company, as we continue to push the boundaries of artificial intelligence (AI) in veterinary medicine.

About the Health to Innovation Hub

The Health Innovation Hub (H2I), housed within the University of Toronto’s Faculty of Medicine, is a collaborative program aimed at fostering the growth of startups and early-stage ventures working in the health and life sciences sectors. The H2I connects innovative companies like ANI.ML Health with world-class resources, mentorship, and access to enable them to transform cutting-edge science into real-world healthcare solutions.

What ANI.ML Health Does

At ANI.ML Health, we specialize in developing AI-driven solutions for veterinary healthcare. Our mission is simple: to empower and transform how veterinary teams diagnose, treat and manage disease while harnessing the power of AI and big data. We expand our knowledge in the form of language modelling, computer vision tasks and spatial computing.

By combining years of veterinary expertise with machine learning expertise and domain-specific models, we will empower veterinarians to deliver more accurate, faster, and personalized care, all while being able to head home on time. Our solutions can analyze massive volumes of medical data—from patient histories and diagnostic images to treatment outcomes

—helping to streamline workflows and ultimately improve the quality of care animals receive.

What’s Next for ANI.ML Health

Through the Health to Innovation Hub, ANI.ML Health will receive mentorship, access to research facilities, and critical connections with key players in the healthcare and AI ecosystems. This partnership will accelerate our R&D efforts and help us bring new AI-based products to the veterinary market even faster.

Our acceptance into the Health to Innovation Hub is a testament to ANI.ML Health’s potential to bring transformational change to the veterinary field. The team recognized our commitment to innovation and our ability to address critical pain points in veterinary medicine through technology. It is both an honour and a privilege to have been accepted into this program and we are incredibly excited. We cannot wait to show you more!

For more details on the Health to Innovation Hub, you can visit their official page. To learn more about ANI.ML Health and our ongoing projects, check out our website.

Link to Voxel51 Blog Post 

Link to talk on YouTube

Healthcare AI & ANIML Health Research

Healthcare AI is fundamentally transforming patient care by improving diagnostic accuracy, optimizing workflows, and enabling personalized treatment plans. Through advanced algorithms and data-driven insights, AI empowers healthcare professionals to make more informed decisions, driving better patient outcomes and operational efficiency. As this transformative technology continues to evolve, it is shaping a future where healthcare becomes more accessible, effective, and equitable for both patients and providers.

Recently, the ANIML Health team had the privilege of presenting to over 100 attendees at the Visual Healthcare AI Meetup, hosted by Voxel51. Voxel51’s platform enables AI research teams to implement, test, and evaluate AI models within clinical contexts, though its capabilities extend far beyond just healthcare applications. This makes it an ideal environment for advancing AI in medicine.

Our research platform focuses on evaluation clinical biomarkers based on real clinical data and translating this back to the clinic and to humans with similar diseases.

At the event, we focused on the growing field of comparative computational oncology, where animal healthcare serves as a platform for developing AI and computational systems that can be translated into the human medical space. This approach also promotes the adoption of AI in veterinary medicine. The benefits of this dual application are clear:

Veterinary medicine is less regulated, providing a unique “playground” for the development and testing of AI models. This allows academic researchers, veterinary teams, and companies to help shape the evolving AI landscape alongside regulators in ways that can inform both human and animal healthcare.

However, the lack of regulation comes with challenges. At ANIML Health, we are deeply committed to transparency because we have observed the rapid adoption of AI platforms in veterinary medicine with little to no reporting on performance metrics or rigorous evaluation of data. This can hinder the long-term success and safety of AI technologies in the field. Ensuring accountability in AI development is critical to maintaining the trust of clinicians, researchers, and, ultimately, the public.

Using computer vision models to identify oral tumours in dogs.

We highlighted several of our projects that we’ve discussed here before, but we are super excited to showcase some preliminary data on oral tumour segmentation modelling using Segment Anything Models (SAM), fine-tuned medical SAMs (medSAM and medSAM2) as well as the development of our own fine-tuned veterinary SAM for oral tumour segmentation.

Pushing this further with segmentation of canine oral tumours.

In the photo above, instead of simply identifying the tumor using computer vision (by placing a box around the affected area), we take it a step further by autosegmenting the precise margins of the tumour. This advanced approach enables both remote monitoring for clinical trials (patient-facing) and volumetric assessments for clinicians (clinician-facing). Additionally, analyzing the textures and pixel-to-pixel relationships enhances our ability to predict tumour histology, behaviour, and treatment response, improving clinical decision-making.

3D oral tumour mapping using iPhone LiDAR photogrammetry

We are equally passionate about extending these techniques to spatial computing. By utilizing iPhone camera acquisition combined with Light Detection and Ranging (LiDAR) sensors, we capture not only images and positional data but also depth information. This technology enables us to create 3D models of patients (and their tumours), which can be used for treatment planning and shared directly with their family members. Additionally, it opens up a wide range of applications, from surgical planning and delineation to the advanced textural analysis and response evaluation mentioned earlier.

Check out the talk links above - we were so fortunate to have the opportunity to discuss what we’ve been up and we can’t wait to show you more!

We’re thrilled to announce that Balto, our consultation platform for veterinary specialists, is officially live!

Developed with the goal of expanding access to remote clinical consultations, Balto connects referring veterinary teams with the specialized support they need to provide the best possible care for their patients.

Getting started is simple—sign up for a FREE account at https://consult.animl.health and start submitting cases to the specialty of your choice. If you need assistance with the sign-up process or submitting a case, you can check out our step-by-step user guide [here].

Our platform offers consultations with specialists in anesthesia, avian & exotic medicine, critical care, surgery & surgical oncology, internal medicine, medical oncology, andd neurology. We’re continually working to onboard more specialists, including experts in cardiology and ophthalmology, to further expand access and support teams around the globe.

Help spread the word! Share Balto with your colleagues on Twitter, LinkedIn, or Instagram and tag us to receive a discount code for your first three consults.

I also need to take a few moments to thank our entire team here at ANI.ML Health. Getting Balto up and running was no small feat and we are so proud of what we’ve accomplished here in such a short time period. There were many sleepless nights and hard work that went into building this platform and I couldn’t be more proud.

We will continue to build features and make improvements as you, our community, continue to provide feedback.

We previously announced our collaboration with Scribenote for ANI.ML Health members! We firmly believe in connecting our community, especially with those who share our values and vision.

As part of this collaboration, our teams will be offering special pricing for users who utilize both platforms. A special discount will be provided for the first 10 consultations for Scribenote users on their ANI.ML Health consults. Veterinarians using our platform will also be eligible for special pricing on Scribenote.

For our specialty consultants, access to Scribenote will be available immediately for cases referred to you. We hope that tools like Scribenote will help improve the efficiency of medical reports and make the process smoother and easier.

Another big shout out to the Scribenote team—we’re really excited to be working together in this space!

That’s all for now! Follow us on social media here or email us to learn more!

-Chris

https://www.johnsnowlabs.com/in-depth-comparison-of-spark-nlp-for-healthcare-and-chatgpt-on-clinical-named-entity-recognition/

As many of you know, our team is deeply passionate about pushing the boundaries of veterinary research. We strive to differentiate ourselves by being transparent and offering as many open-source tools as possible. We also want to assure the veterinary community of our commitment to publication, so you can see firsthand what we’ve been working on.

Last month, we submitted an article to Veterinary & Comparative Oncology that evaluates the use of an open-source Named Entity Recognizer (NER). Named entity recognition is a form of language modelling that categorizes entities (words or strings of words) into groups such as diagnoses or medications. This enables the development of tools that can effectively analyze medical text and provide context to words or sentences. The tool we evaluated, called BioEpiNER, was originally developed by a group at the University of Toronto. Our research explores whether these tools, initially trained on human medical text, can be applied to veterinary text. Our paper is still under review, so I won’t reveal the results just yet, but we hope it will have a significant impact!

We are also continuing our collaboration with both the University of Guelph and the team at ScanLand, focusing on 3D imaging of oral tumours using LiDAR technology—straight from your iPhone! We aim to extend this technology into an application for clinicians and patient families for remote clinical trial monitoring, along with other machine learning-based tools for clinicians, which we’ll be sharing soon.

3D iPhone LiDAR scan of a patient with an oral tumour of the rostral maxilla

We’re really excited for this collaboration!

We are thrilled to announce a collaboration with Scribenote for ANI.ML Health members! We firmly believe in connecting our community, especially with those who share our values and vision.

As part of this collaboration, our teams will be offering special pricing for users who utilize both platforms. A special discount will be provided for the first 10 consultations for Scribenote users on their ANI.ML Health consults. Veterinarians using our platform will also be eligible for special pricing on Scribenote.

For our specialty consultants, access to Scribenote will be available immediately for cases referred to you. We hope that tools like Scribenote will help improve the efficiency of medical reports and make the process smoother and easier.

A big shout out to the Scribenote team—we’re really excited to be working together in this space!

Our team is working tirelessly this week to finalize the many features we promised. Notification routing and handling of revisit consultations, both for the end user and specialist, are our top priorities. We’ll be entering the testing phase by the end of this week in preparation for launch 🚀.

We also updated our website this week to include a user walkthrough for veterinary teams. If you want a sneak peek at what things look like and how a typical consultation workflow goes, check it out here.

That’s all I can say for now, but stay tuned for more!

-Chris

It’s finally time to get it out there.

Our team has been in stealth mode for just under a year now, so I thought it was a good time to finally start talking about what we’ve been up to. As a veterinary oncologist, and having been immersed in the veterinary profession since I was 14 years old, I’ve had the opportunity to see and work in just about every facet of the profession. From chicken barns for research, to small animal clinics, to specialty hospitals and research institutions, this diverse experience has empowered our team to critically evaluate what veterinary teams need in both the short-term and long-term.

Our mandate was developed to serve three primary goals:

1. Veterinary teams deserve tools actually built by veterinarians that they can trust and contribute to.

2. Veterinary teams deserve tools that will improve their quality of life.

3. We will always build tools that help improve patient care and patient service.

In 2023, as a clinician, I was frustrated. My administrative time far surpassed my clinical time, and I was feeling burnout more than ever. Fortunately, our team was formed behind the scenes with the goal to empower clinicians to do better work, faster work, and really focus on what we trained to do—provide care for animals.

Our profession has changed dramatically over the last few years, with some changes for the better and some certainly for the worse. We felt it was time to put the reins back in the hands of veterinary teams and make a real clinical impact. At ANI.ML, we are dedicated to developing AI-based solutions that not only enhance the efficiency and effectiveness of veterinary practices but also ensure that veterinarians can focus on what truly matters: delivering high-quality care to their patients.

We look forward to sharing more about our journey, our innovations, and how we aim to transform the veterinary profession. Stay tuned as we continue to work towards creating a brighter future for veterinary medicine!

The medical consulting platform. What is it?

To help empower veterinary teams, we first needed to acknowledge the shortage of veterinarians, particularly veterinary specialists. The corporatization of veterinary medicine has had a dilutional effect, and given years of poor intern and resident salaries and quality of life, we are seeing a decline in new veterinary specialists each year. Access to specialty care is typically concentrated in high-density cities where corporate institutions continue to build and expand. We wanted to provide remote and rapid access to care for veterinary teams so you never feel alone, leading to the creation of our consulting platform.

The platform is simple to use. You create an account with ANI.ML, select the specialty you want to engage with and then the specific specialist. Upload your referral documents and answer a brief list of questions, and you're done. Within 24 hours of receipt, you’ll receive a response from one of our specialists. If you need to revisit the same case as things change, that’s no problem either.

Sign in / Sign up page for the platform

For specialists, we recognized that unless you worked for larger labs or corporations with internal consultation platforms, there was no way for veterinary specialists to act independently and consult remotely (except through virtual health or phone calls). We wanted to create a platform that allows veterinary specialists to work from anywhere in the world, at any time, and provide remote care to teams in need. This also offers specialists the unique opportunity to represent their respective practices and enhance their relationships with primary care veterinarians.

The future.

We have several exciting products currently in development, and I’m thrilled to share them with you soon. Like any innovative venture, these things take time. However, we are committed to transparency throughout the entire process, from development to deployment. Moving forward, we’ll be sharing our entire journey with you, the veterinary community, so you know exactly what we’re building and how we’re building it. It’s important to us that you understand how our products work, the research we are engaged in, and our mission to transform the veterinary profession.

To see what we’re up to, please follow our public GitHub repo available here: https://github.com/ANI-ML

We’ll keep you updated on our research efforts, publications, and more exciting news.

We invite you to join us on this journey. Share our progress with your friends and colleagues, and stay connected with us on social media to follow our updates and innovations!

Follow us on:
LinkedIn: https://www.linkedin.com/company/animlhealth/

Twitter: https://x.com/ANIML_Health

Instagram: https://www.instagram.com/animl_health/

DeepTag: inferring diagnoses from veterinary clinical notes

Allen Nie, Ashley Zehnder, Rodney L Page, Yuhui Zhang, Arturo Lopez Pineda, Manuel A Rivas, Carlos D Bustamante, James Zou

The study presents the development of DeepTag, an algorithm designed to automatically infer diagnostic codes from veterinary clinical notes, aiming to alleviate the lack of standardized coding infrastructure in veterinary medicine. DeepTag utilizes a bidirectional long-short-term memory network (BLSTM) for training, which extends the multitask LSTM with a hierarchical objective to capture the semantic structures between diseases. It is trained on a dataset of 112,558 manually annotated veterinary notes, enabling automated disease annotation across a broad range of clinical diagnoses with minimal preprocessing. The researchers also address the challenges of cross-hospital coding tasks and the differences in text style between academic and private practice settings.

The paper highlights the scarcity of structured coding and standardized nomenclatures in veterinary medicine, hindering clinical research and public health monitoring efforts. It also emphasizes the potential translational impact of utilizing spontaneous disease models in animals for the study of human diseases, particularly in noninfectious diseases and drug development pipelines. Furthermore, the study emphasizes the need for natural language processing (NLP) tools in the veterinary community to convert free-text clinical notes into structured information, given the abundance of clinical summaries stored as electronic health records in various hospitals and clinics.

Overcoming Challenges

The research team worked on addressing challenges related to the generalization of a tagging system across different academic and private practice datasets, as well as training the system to abstain from making predictions when uncertain. The study proposes novel methods for applying broad disease codes to clinical records and demonstrates the potential implications of such algorithms for real-world implementations, highlighting the significant research required to optimize methods for domain adaptation, as well as the need for expert judgment to improve the overall

workflow.

DeepTag Algorithm Performance and Future Directions

The DeepTag algorithm was found to outperform baseline models on an external private practice dataset, demonstrating improvements in performance and precision, particularly with learned abstention rules. The study concludes with discussions on future directions, such as leveraging unsupervised representation learning and considering non-disorder disease codes in clinical records for further refinement of automated coding systems in veterinary medicine.

Link to Article: DeepTag: inferring diagnoses from veterinary clinical notes - PubMed (nih.gov)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

Analysing the Applicability of ChatGPT, Bard, and Bing to Generate Reasoning-Based Multiple-Choice Questions in Medical Physiology

Mayank Agarwal, Priyanka Sharma, Ayan Goswami

Introduction

The research paper introduces the concept of artificial intelligence (AI) as the development of computer systems that mimic human cognitive function to solve complex problems. AI is rapidly advancing in various scientific fields and particularly in healthcare and biomedical research. The study focuses on three AI models: ChatGPT, Bard, and Bing, which generated 110 multiple-choice questions (MCQs) on June 2, 2023. The MCQs were rated by two physiologists on three parameters: validity, difficulty, and reasoning ability using a scale ranging from 0 to 3. The ratings were recorded in an Excel spreadsheet for subsequent analysis. This research sheds light on the cognitive abilities exhibited by AI models and their potential implications in different domains, including healthcare and scientific research.

Parameter Rating

The Parameter Rating section of the research paper confirms that the multiple-choice questions (MCQs) generated by the AI models for the medical physiology subject are deemed valid and clear without ambiguity. The physiologists, who were blinded to the AI model origins of the MCQs, conducted the analysis and provided ratings for validity, difficulty, and reasoning ability. After a week of evaluation, the ratings from the two physiologists were averaged, and the results were compiled for further statistical analysis. The originality of the content was also checked using Turnitin software. Moreover, a figure in the paper outlines the method used in the study, indicating the rigorous process followed. This section elucidates the systematic approach taken to assess the quality and suitability of the AI-generated MCQs for medical physiology competence.

Statistical Analysis

The researchers inputted the data into Microsoft Excel 365 and conducted statistical analysis using IBM SPSS Statistics Version 27.0 for Windows, opting for non-parametric tests due to the ordinal nature of the data. They presented the data using median and interquartile range (Q1-Q3). To compare distribution across total and module-wise responses, they employed the independent sample Kruskal-Wallis test and conducted post-hoc tests for pairwise comparisons. Additionally, they evaluated the agreement in s

cores between two raters using Cohen's Kappa (Κ) and considered a p-value <0.05 to indicate statistical significance.

Results

The study evaluated the performance of ChatGPT, Bard, and Bing in generating multiple-choice questions (MCQs) for the NMC CBME curriculum. ChatGPT and Bard produced 110 MCQs for 22 competencies, while Bing provided only 100 MCQs, failing for two competencies. Bing received '0' ratings for validity, difficulty, and reasoning ability for these competencies. Overall, Bing generated the least valid MCQs, while ChatGPT generated the least difficult ones. Turnitin test similarity indices showed 39% for ChatGPT, 49% for Bard, and 52% for Bing. Inter-rater reliability was strong (Cohen's K ≥ 0.8) for all AIs and parameters. These results highlight differences in the performance of AI models in generating MCQs and emphasize the need for further development to improve validity and difficulty of questions.

The analysis of 110 multiple-choice questions (MCQs) generated by AI models such as ChatGPT, Bard, and Bing revealed interesting findings. ChatGPT appeared to be the slowest in generating MCQs, although specific timings were not recorded. Bard produced 47 MCQs with the stem containing the suffix "Which of the following is the most important," and 54 MCQs had "all of the above" as an option. Moreover, Bard was the only AI that provided answers to the MCQs with an explanation. On the other hand, all the generated MCQs by Bing consistently had the suffix "Which of the following is not," indicating a negative verb construction. These insights provide valuable information on the performance and characteristics of the AI models in generating MCQs, which can have implications for their application in various fields including healthcare and biomedical research.

Discussion

The study evaluated the ability of AI models ChatGPT, Bard, and Bing in creating valid, difficult, and reasoning-based multiple-choice questions (MCQs) in medical physiology. ChatGPT generated the most valid MCQs but the least difficult ones, while all AIs struggled to generate high-level reasoning-based MCQs. Shortcomings in MCQ generation were identified, such as the presence of negative words and non-ideal question structures. Variations in performance across different competency modules and text similarity indexes were observed among the AIs. Despite AI's potential in medical education, challenges in ensuring accuracy and reliability remain. Previous studies have shown the successful use of ChatGPT in medical education and its capacity to provide accurate responses, but its abilities in creating reasoning-based MCQs lag behind human intelligence. The integration of AI into medical education is seen as inevitable and could revolutionize the learning experience, although the accuracy and reliability of information provided by AI systems remain a significant challenge.

Limitations

The research study had several limitations. It focused solely on assessing the ability of AI systems to generate multiple-choice questions (MCQs) related to medical physiology, which may limit the generalizability of the findings to other subjects or domains. Another limitation was the reliance on a single user for conversational interactions with ChatGPT, potentially leading to varied responses with different users or at different times. The study also recognized that paraphrasing questions could introduce variations in ChatGPT responses, impacting its overall performance evaluation. Additionally, the subjective scoring of AI-generated responses relied on human evaluators, potentially introducing evaluation bias despite efforts to mitigate it. Lastly, the study did not include MBBS students in the MCQ item analysis, presenting a limitation in the evaluation process.

Link to Article: Analysing the Applicability of ChatGPT, Bard, and Bing to Generate Reasoning-Based Multiple-Choice Questions in Medical Physiology - PubMed (nih.gov)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

The Future Ethics of Artificial Intelligence in Medicine: Making Sense of Collaborative Models

Torbjørn Gundersen, Kristine Bærøe

The article explores the ethical challenges and concerns arising from the use of artificial intelligence (AI) and machine learning in medical decision-making. It highlights that while deep learning technologies have the potential to improve diagnostic procedures and treatments, they also pose significant ethical problems such as the risk of error, lack of transparency, and disruption of accountability. The paper discusses the role of medical doctors, AI designers, and other stakeholders in making AI ethically acceptable and proposes four models for integrating their input: the ordinary evidence model, the ethical design model, the collaborative model, and the public deliberation model.

The ordinary evidence model suggests that the responsible use of AI in medicine is ensured by the application of established medical expertise and ethical principles by medical doctors. However, this model faces objections, such as the difficulty of attributing accountability and the opacity of AI. The ethical design model proposes encoding ethical values directly into algorithms to make them ethically acceptable. While this approach takes ethical considerations into account in design, it may lead to a lack of fit between design and medical practice and overlook the need for ethical deliberation.

The collaborative model emphasizes collaboration and mutual engagement between AI designers, bioethicists, and medical doctors to align algorithms with medical expertise and ethical principles. It suggests that meaningful communication and collaboration are crucial in the design and use of medical AI. Furthermore, the public deliberation model involves broad public debate on the benefits and costs of AI in medicine, beyond the involvement of AI designers, medical experts, and bioethicists.

Overall, the article underscores the need for collaboration and ethical deliberation in the design and use of AI in medicine, and it offers a systematic discussion of different models for addressing the ethical concerns posed by medical AI.

Link to Article: The Future Ethics of Artificial Intelligence in Medicine: Making Sense of Collaborative Models - PubMed (nih.gov)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

Feasibility study of computed tomography texture analysis for evaluation of canine primary adrenal gland tumors

Sibylle Maria Kneissl, Silvia Burti, Kyungsoo Lee, Jinhyong Goh, Jaeyoung Jang, Jeongyeon Hwang, Jungmin Kwak, Jaehwan Kim, Kidong Eom

Introduction

Canine primary adrenal gland tumors are rare but can be classified by origin, behavior, and functionality. Adenoma and adenocarcinoma are common, followed by pheochromocytoma. CT plays a crucial role in preoperative tumor diagnosis, but has limitations in distinguishing tumor types. Texture analysis is a promising quantitative method for evaluating tumor heterogeneity. It has been used in human medicine but not extensively in veterinary medicine, particularly in the differentiation of primary adrenal gland masses. This study aimed to characterize the texture features of canine primary adrenal gland tumors using CT and assess its diagnostic performance using the ROC curve. The classification of tumors by origin and functionality, as well as the potential of CT texture analysis as a method for differentiating malignant from benign adrenal gland tumors, were investigated. The study underscores the need for further investigation of CT texture analysis in canine adrenal gland tumors.

CT image acquisition

The CT images for the study were acquired from various multi-detector CT scanners using standardized abdominal imaging protocols. The specific scanner models, slice thickness, tube voltage (kVP), and mAs are detailed in the Supplementary Table. Postcontrast images were obtained either using the bolus tracking technique or at specific time intervals after the injection of the contrast medium, including arterial phase (∼20 s), portal phase (∼40 s), and delayed phase (70 s to 2 min post-injection). Due to the retrospective nature of the study, detailed protocols such as the manufacturer of the contrast medium, dosage, injection rate, and specific scanner parameters were not available for analysis.

Analysis

The study involved retrospective review of CT images of canine adrenal gland tumors by a single investigator using Digital Imaging and Communications in Medicine (DICOM) viewing software. The evaluation, supervised by a professor, was blinded to the histopathological diagnosis. Measurements included maximal diameter on short and long axes, and mean, maximum, and minimum attenuation values in Hounsfield units (HU). Circular or ovoid regions of interest (ROIs) were drawn on the transverse plane, excluding calcification and macroscopic fat. ROIs were placed consistently on pre-contrast and post-contrast images, and the difference in mean attenuation value on pre-contrast images was calculated on each post-contrast image. This conventional CT analysis aimed to characterize the tumors. The methodology employed in this study involved meticulous and comprehensive quantitative assessment of CT images, providing important insights into the characterization of canine adrenal gland tumors.

Qualitative features

The qualitative features evaluated in the study included location (right or left), shape (round, oval, lobulated), mass contour (smooth, irregular), contrast enhancement type (homogeneous, heterogeneous), pattern (stable, progressive, washout), degree of enhancement (none, minimal, mild, moderate, intense), rim enhancement, intratumoral calcification, and adhesion or invasion of adjacent vessels. The enhancement pattern was categorized as stable (minimal enhancement), progressive (gradual increase), or washout (peak followed by reduction). Degree of enhancement was assessed using mean HU differences. A 7-point CT grading system was used to evaluate vessel adhesion or invasion, with higher grades indicating a stronger possibility of invasion. These qualitative features were instrumental in characterizing canine adrenal gland tumors and assessing their potential malignancy, aiding in the differentiation between benign and malignant tumors based on their CT features.

Lesion segmentation and texture analysis

The study performed texture analysis on canine adrenal lesions using pre-contrast and delayed-phase CT images with 3D Slicer software. Eighteen out of 25 dogs were included due to variations in CT scanners and protocols that couldn't be harmonized using ComBat method. The method proved effective in removing batch effects while preserving texture characteristics. Semi-automatic segmentation and feature extraction were conducted on the adrenal lesions, extracting 18 first-order and 75 second-order statistics using various matrices. The process was repeated by a single investigator, blinded to histopathologic results, and supervised by a radiologist. Additionally, the ComBat harmonization method was applied to the extracted texture features. This technique addresses the interscanner variability in radiomic features and shows promise for harmonizing data in multicenter radiomic studies. However, it's essential to consider the consistency of image acquisition and reconstruction for reproducibility of radiomic features.

Statistical analysis

In the statistical analysis of the study, qualitative and quantitative CT features were evaluated using various tests. Qualitative features were assessed using the Fisher exact test, while quantitative features underwent normality testing and one-way analysis of variance to compare differences between tumor types. The intraobserver intraclass correlation coefficients for radiomic features were >0.90, indicating good consistency in feature extraction. Radiomic features were compared between tumor types using the Kruskal-Wallis H test. The diagnostic performance of differentiating tumor types was evaluated using the area under the ROC curve (AUC), with values classified as fail, poor, fair, good, or excellent. The maximum Youden index was used to determine the cut-off value for features with the highest AUC. Diagnostic parameters such as sensitivity, specificity, PPV, and NPV were obtained using the cut-off value. MedCalc and SPSS software were used for statistical analyses, with a significance level of p<0.05. Overall, the histopathological diagnosis served as the gold standard for reference in the study's statistical analyses.

Results

The study included a total of 40 dogs, with 25 meeting the inclusion criteria. Exclusions were made for cases without histopathological results, benign lesions, loss of CT data, and adrenal mass rupture. Of the 25 included dogs, 18 were involved in the texture analysis. Tumors were diagnosed as adrenocortical adenoma (AA) in 48% of cases, adrenocortical carcinoma (ACC) in 28% of cases, and pheochromocytoma (PHEO) in 24% of cases, based on surgical excisional biopsy findings. Some cases of AA were found to have concurrent nodular hyperplasia and extramedullary hematopoiesis. Capsular invasion was confirmed in only one of the PHEOs. The study involved various dog breeds, with no significant differences in age and sex between the tumor types.

Quantitative Features

The study found that in conventional CT evaluation of primary adrenal gland tumors, the mean and maximum Hounsfield units (HUmean and HUmax) on pre-contrast images were significantly higher in pheochromocytomas (PHEO) than in adrenocortical carcinomas (ACC). No significant differences were observed between adrenocortical adenomas (AA) and ACC, or between AA and PHEO. The HUmax in the arterial phase was notably higher in PHEO compared to other neoplasms, although this difference did not reach statistical significance. The study also noted that the mean HU difference and maximum diameter did not show significant correlation with tumor type. Additionally, ACC exhibited the smallest size among the three tumor types. Overall, the study's findings on the quantitative CT features of adrenal gland tumors highlight distinct characteristics that may aid in their differentiation, particularly between PHEO and ACC.

Qualitative features

The qualitative features of canine primary adrenal gland tumors were evaluated using conventional CT evaluation. The study found that none of the features were significantly associated with the tumor type. In the AA group, all cases showed smooth tumor margins, but the difference was not statistically significant between the tumor types. No direct vessel invasion or compression was observed in any case. Most tumors had a moderate possibility of tumor vascular invasion or adhesion. Additionally, the majority of tumors in the pre-contrast phase and all tumors in the portal and delayed phases showed heterogeneous enhancement. The washout pattern of contrast enhancement was identified in only 1 case, and progressive patterns were most frequently observed regardless of the tumor type. These findings provide insights into the qualitative features of canine adrenal gland tumors, with implications for their classification by origin and functionality using CT evaluation.

CT texture analysis

The study compared CT texture features of primary adrenal gland tumors to differentiate tumor types and identify potential malignancy. Eight second-order statistics from pre-contrast images showed significant differences between tumor types, while no significant features were identified in the delayed phase. Features such as correlation, maximal correlation coefficient, and gray-level non-uniformity normalized were notable for showing significant differences between adrenocortical adenoma (AA) and adrenocortical carcinoma (ACC), as well as between ACC and pheochromocytoma (PHEO). However, no significant differences were found between AA and PHEO. The study demonstrated the potential of CT texture analysis in differentiating tumor types, especially between AA, ACC, and PHEO. These findings suggest that CT texture analysis may be a promising method for characterizing primary adrenal gland tumors and distinguishing malignant from benign tumors, particularly in the pre-contrast phase.

Evaluation of diagnostic performance

The research paper evaluated the diagnostic performance of CT texture features for differentiating adrenal gland tumors. The HUmean and HUmax on the pre-contrast image demonstrated high area under the curve (AUC) values for diagnosing adrenocortical carcinoma (ACC) and pheochromocytoma (PHEO). The AUC values for distinguishing ACC from other adrenal gland tumors were high, with the maximum correlation coefficient (MCC) showing the highest AUC (0.969). Additionally, radiomic features exhibited AUC values indicative of fair to excellent diagnostic performance for each tumor type. Specifically, run entropy showed the highest AUC (0.787) for discriminating adrenal adenoma (AA), while short‐long‐run high‐gray‐level emphasis (SLDGLE) showed the highest AUC (0.889) for discriminating PHEO. The study also provided cutoff values for differentiating ACC and PHEO, along with the radiomic features' corresponding AUC values. These findings demonstrate the potential of CT texture analysis in accurately characterizing and differentiating adrenal gland tumors.

Discussion

The study evaluated the feasibility of using CT texture analysis to differentiate adrenal gland tumors, including cortical AA, ACC, and PHEO. It identified several texture features on pre-contrast images with statistical significance, showing the potential of CT texture analysis to distinguish adrenal gland tumors. Notably, the study found significant differences in texture features between tumor types, particularly ACC and PHEO, indicating the potential of CT texture analysis in canine adrenal glands using pre-contrast imaging. The study acknowledged some limitations, including the small sample size and the need for further validation and model establishment in larger prospective studies. Overall, the findings suggest that CT texture analysis may be useful for distinguishing adrenal gland tumor types and aiding in clinical decision-making, highlighting the potential of this approach in canine primary adrenal gland tumor characterization.

Link to Article: Frontiers | Feasibility study of computed tomography texture analysis for evaluation of canine primary adrenal gland tumors (frontiersin.org)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

MEDAGENTS: Large Language Models as Collaborators for Zero-shot Medical Reasoning

Xiangru Tang , Anni Zou, Zhuosheng Zhang, Yilun Zhao, Xingyao Zhang, Arman Cohan, Mark Gerstein

This paper discusses the challenges faced by large language models (LLMs) in adapting to the medical field and proposes a multi-disciplinary collaboration (MC) framework to address these challenges. The framework leverages role-playing LLM-based agents to participate in collaborative multi-round discussions to enhance LLM proficiency and reasoning capabilities. The training-free and interpretable framework encompasses gathering domain experts, proposing individual analyses, summarizing these analyses into a report, iterating over discussions until a consensus is reached, and making a decision.

Discussion of the MC Framework

The paper focuses on the zero-shot scenario and shows that the proposed MC framework excels at mining and harnessing medical expertise in LLMs and extends its reasoning abilities. The research also evaluates the framework's performance on medical question-answering tasks and identifies four common error types through human evaluation. The proposed MC framework outperforms zero-shot baselines and demonstrates comparable performance with a strong few-shot baseline.

Evaluation of the MC Framework

The paper also highlights the challenges of obtaining high-quality instruction-tuning data in the medical domain and the limitations of existing instruction-tuning methods. It emphasizes the success of LLM-based agents in multi-agent collaboration, which brings the model's embedded expertise to the fore and enhances its reasoning capabilities over multi-round interactions.

Conclusions

In conclusion, the paper's major contributions include proposing a novel multi-disciplinary collaboration framework for question-answering tasks in the medical domain, presenting experimental results demonstrating the effectiveness of the MC framework, identifying and categorizing common error types through human evaluation, and shedding light on potential future studies to address the identified limitations and enhance the framework's proficiency and reliability.

Link to Article: [2311.10537] MedAgents: Large Language Models as Collaborators for Zero-shot Medical Reasoning (arxiv.org)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

Machine learning predicts histologic type and grade of canine gliomas based on MRI texture analysis

Pablo Barge, Anna Oevermann, Arianna Maiolini, Alexane Durand

Introduction

Texture Analysis (TA) mathematically extracts quantitative information from medical images and provides objective measurements of tumor heterogeneity. Texture features can be used as input for machine learning (ML) models, which have shown promise in differentiating various pathologies in small animal medicine. However, previous attempts to use TA-based ML classifiers for canine gliomas have been unsuccessful. The study aims to overcome these limitations by using multiple segments across the entire tumor volume, which has not been done in canine patients before. The researchers hypothesize that this approach, combined with different ML classifiers, will lead to a more accurate classification of glioma types and grades based on MRI-TA compared to conventional MRI.

Materials and Methods

In this study, the researchers used neuroimaging techniques to analyze brain tumors. The study design was presented in a figure, and images were converted into a specific format. Different types of brain images were aligned using software, and the tumor's enhancing and non-enhancing parts, as well as the perilesional vasogenic edema, were manually segmented by two experts. The enhancing segment referred to areas of the tumor that showed contrast enhancement, while the non-enhancing segment represented tumor areas without enhancement. The vasogenic edema segment included hyperintense areas in the peri-tumoral white matter. Certain regions, such as meninges, large vessels, intratumoral cysts with suppressing FLAIR signals, and areas affected by partial volume averaging artifact, were not included in any of the segments. The segmentations were then exported as label-maps to extract texture features for further analysis.

Pre-processing techniques and texture feature extraction were carried out using a freeware software called LIFEx. The goal of pre-processing was to standardize the images in terms of pixel spacing, gray-level intensities, and gray-level histogram bins. Image interpolation was done to ensure a consistent in-plane resolution of 1x1 mm while preserving the original slice thickness. Image discretization involved using a fixed bin number of 32, and the gray levels of the MRI within the segmentations were normalized to the mean plus or minus three standard deviations. A total of 61 texture features were extracted per segment in each sequence, resulting in a maximum of 732 texture values per tumor.

Supervised ML algorithms such as Support Vector Machine (SVM), Random Forest (RF), and k-nearest neighbors (kNN) were used for classification and regression problems in medical imaging. The models were programmed using Python and various libraries and run on Google Colab. Before feeding the models, the data was normalized using z-score normalization and dimensionality reduction was performed using Principal Component Analysis (PCA). A leave-one-out cross-validation technique was used to assess the performance of the models. Different datasets were prepared, including all extracted texture features and smaller datasets with specific segments and sequences. The models aimed to predict glioma histological type and grade, and the predictions were compared to histopathological diagnosis to calculate accuracy, sensitivity, specificity, and AUC. The study included a total of 13 datasets and excluded oligosarcomas from the multiclass models due to the low number of cases.

Results

Machine learning (ML) classifiers were used to predict different types of tumors in brain imaging data. The multiclass ML classifiers had an average accuracy of 77% for predicting tumor types, with varying sensitivity and specificity values for different tumor types. The binary ML classifiers had an average accuracy of 76% for predicting high-grade gliomas, with better performance from SVM and kNN classifiers than RF. The ML models were not fed with the enhancing segment as only a few cases showed enhancement. As a result, further analysis was conducted on a smaller dataset. The SVM multiclass classifier performed well for predicting tumor types on edema segments, while the SVM and kNN binary classifiers achieved the best performance for predicting high-grade gliomas on non-enhancing segments in T2w and T1w postcontrast sequences. Overall, the classifiers had varied performances on different sequences and segments for predicting tumor types and high-grade gliomas.

Discussion

The researchers found that TA-based ML models using multiple segments of the tumor volume provided more accurate classification compared to conventional MRI. The SVM and kNN models demonstrated higher accuracy, sensitivity, and specificity in distinguishing between high-grade and low-grade gliomas compared to conventional MRI. The study also highlighted the potential of TA-based ML in improving treatment planning and prognosis assessment for canine gliomas. However, there were limitations such as the imbalanced dataset and difficulty in differentiating between glioma types. The study suggests the need for further research to explore the different texture metrics across glioma types and grades. The use of automated techniques for tumor segmentation in dogs is also recommended for future studies.

Conclusion

The authors conclude that machine learning models using MRI-TA can effectively distinguish different types and grades of intracranial canine gliomas. They found that the support vector machine (SVM) model had accuracies of up to 94% for tumor type discrimination and 87% for tumor grade discrimination. The texture features that were most useful for differentiating tumor types were associated with peri-tumoral edema in T1w images, while those for tumor grades were related to the non-enhancing part of the tumor in T2w images. However, the authors recommend conducting further multicenter studies with larger sample sizes to validate and strengthen these findings.

Link to Article: Machine learning predicts histologic type and grade of canine gliomas based on MRI texture analysis - PubMed (nih.gov)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

Evaluation of supervised machinelearning algorithms to distinguish between inflammatory bowel disease and alimentary lymphoma in cats

Abdullah Awaysheh, Jeffrey Wilcke, François Elvinger, Loren Rees, Weiguo Fan, Kurt L. Zimmerman

This article explores the use of machine-learning algorithms to differentiate between inflammatory bowel disease (IBD) and alimentary lymphoma (ALA) in cats.

The authors developed three prediction models using three different machine-learning algorithms: naive Bayes, decision trees, and artificial neural networks. The models were trained and tested using data from complete blood count (CBC) and serum chemistry (SC) results from normal cats, cats with IBD, and cats with ALA. The naive Bayes and artificial neural networks models achieved higher classification accuracy compared to the decision tree model.

The area under the receiver-operating characteristic curve for classifying cases into the three categories was 83% for naive Bayes, 79% for decision tree, and 82% for artificial neural networks. The models were able to provide another noninvasive diagnostic tool to assist clinicians in differentiating between IBD and ALA, as well as between diseased and nondiseased cats.

The models used a subset of CBC and SC variables to classify the cats, with the naive Bayes classifier using 10 variables and the artificial neural networks classifier using 4 variables. The authors suggest that the naive Bayes and artificial neural networks classifiers are the best choices for constructing prediction models in this specific use case.

Overall, this study demonstrates the potential of machine-learning algorithms as noninvasive diagnostic tools for differentiating between gastrointestinal diseases in cats.

Link to Article: Evaluation of supervised machine-learning algorithms to distinguish between inflammatory bowel disease and alimentary lymphoma in cats - PubMed (nih.gov)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.

AlphaFold accelerates artificial intelligence powered drug discovery: efficient discovery of a novel CDK20 small molecule inhibitor

Feng Ren, Xiao Ding, Min Zheng, Mikhail Korzinkin, Xin Cai, Wei Zhu, Alexey Mantsyzov, Alex Aliper, Vladimir Aladinskiy, Zhongying Cao, Shanshan Kong, Xi Long, Bonnie Hei Man Liu, Yingtao Liu, Vladimir Naumov, Anastasia Shneyderman, Ivan V. Ozerov, Ju Wang, Frank W. Pun, Alan Aspuru-Guzik, Michael Levitt, Alex Zhavoronkov

Introduction to AlphaFold and AI-powered drug discovery

In this scientific article, researchers demonstrate the use of the AlphaFold computer program to accelerate artificial intelligence (AI)-powered drug discovery. AlphaFold, developed by DeepMind, has the ability to predict protein structures. The researchers applied AlphaFold to their AI-powered drug discovery platforms, PandaOmics and Chemistry42, to identify a small molecule inhibitor for cyclin-dependent kinase 20 (CDK20), a protein associated with hepatocellular carcinoma (HCC), a type of liver cancer.

Analysis and synthesis of compounds using PandaOmics and Chemistry42

Using PandaOmics, the researchers analyzed text and OMICs data from multiple datasets to identify potential targets for HCC. CDK20 was selected as the target due to its strong association with HCC and limited structural information. Chemistry42 was then used to generate molecules based on the predicted CDK20 structure from AlphaFold. From the pool of generated molecules, seven compounds were synthesized and tested. One compound,

ISM042-2-001, demonstrated binding affinity to CDK20 with a Kd value of 9.2 ± 0.5 mM.

Successful application of AlphaFold predictions in drug discovery and future prospects

This work demonstrates the successful application of AlphaFold predictions in the drug discovery process, particularly for identifying hit molecules against novel targets with limited structural information. The integration of AI-powered platforms PandaOmics and Chemistry42 enabled the rapid discovery of a CDK20 inhibitor within 30 days. Further optimization and evaluation of the hit molecule are ongoing. This study highlights the potential of AI and protein structure prediction for accelerating drug discovery.

Link to Article: [2201.09647] AlphaFold Accelerates Artificial Intelligence Powered Drug Discovery: Efficient Discovery of a Novel Cyclin-dependent Kinase 20 (CDK20) Small Molecule Inhibitor (arxiv.org)

This article was summarized by an AI tool that uses natural language processing. The tool is not perfect and may make mistakes or produce inaccurate or irrelevant information, but is reviewed by the post’s author prior to publishing. If you want to learn more about the article, please refer to the original source that is cited at the end of the article.