We just wrapped up the September ‘24 Visual AI in Healthcare Meetup, and if you missed it or want to revisit it, here’s a recap! In this blog post you’ll find the playback recordings, highlights from the presentations and Q&A, as well as the upcoming Meetup schedule so that you can join us at a future event.
Interpretable AI Models in Radiology
AI methods have reached and even surpassed human-level accuracy in numerous areas of healthcare. However, adoption of these technologies into clinical workflows, where interpretability is of paramount importance, is slower compared to other industries. In this talk, we will present an overview of our research in improving the interpretability of AI models in medical image analysis through counterfactual examples and radiologist gaze data collection.
Speaker: Dr. Tasdizen is a Professor in Electrical and Computer Engineering and the Scientific Computing and Imaging (SCI) Institute at the University of Utah. His areas of expertise are medical image analysis and machine learning.
Q&A
- Can we develop AI model with immersive technology that are inherently interpretable
- What are some of the next steps you’re exploring to further improve interpretability using radiologist gaze data?
- Do you think providing the localization in the radiograph is good enough for the interpretability of the model?
- Have you considered using a mouse and having the radiologist explaining their decision as an alternative ground truth?
- In what ways do radiologists most often fail to correctly assess chest x-rays?
- Why are we interested in the radiologist gaze if it’s seen that AI tends to outperform them?
- Who is legally liable for mistakes, and does malpractice insurance cover this?
Resource Links
Bridging Species with Pixels: Advancing Comparative Computational AI in Veterinary Oncology
Roughly 50% of dogs over the age of 10 years will develop cancer. Animals are now part of the family, and veterinary medical care now approximates what is available in humans. We are now at a pivotal time where AI platforms and products can expedite clinical discovery and decision – making and accelerate innovation. In this talk, we will provide a high-level overview of comparative AI and the work our team has initiated to evaluate both radiomic and language-based models in veterinary medicine.
Speakers: Dr. Christopher Pinard, DVM DVSc DACVIM (Oncology) is the CEO and co-founder of ANI.ML Health Inc., an adjunct professor in the Department of Clinical Studies at the Ontario Veterinary College, University of Guelph, a Medical Oncologist at Lakeshore Animal Health Partners, a Research Fellow at Sunnybrook Research Institute, and a Faculty Affiliate with the Centre for Advancing Responsible and Ethical Artificial Intelligence (CARE-AI) at the University of Guelph. His research focuses on comparative computational oncology and the development of computer vision and language model-based tools for clinical applications. Dr. Kuan-Chuen Wu builds A.I. products and Engineering solutions via scientific research, technological development, and global teaching. With a Harvard-Stanford education in multi-disciplinary engineering, data science, and business management, he leads multi-functional teams and communities in generative A.I. and predictive A.I. using hardware, software, theory plus ingenuity for societal good.
Q&A
- Are there any studies which show the merit of pre-training on human data?
- Why did you select YOLOV2 vs the latest one, YOLOV8?
- The iPhone can use other sensors in addition to photos for 3D reconstruction, have you considered this?
Resource Links
- Learn more about ANI.ML Health
- Dr Pinard’s research publications
Deep-Dive: NVIDIA’s VISTA-3D and MedSAM-2 Medical Imaging Models
In this talk, we’ll explore two medical imaging models. First, we’ll dive into NVIDIA’s Versatile Imaging SegmenTation and Annotation (VISTA) model which combines semantic segmentation with interactivity, offering high accuracy and adaptability across diverse anatomical areas for medical imaging. Finally, we’ll explore MedSAM-2, an advanced segmentation model that utilizes Meta’s SAM 2 framework to address both 2D and 3D medical image segmentation tasks.
Speaker: Daniel Gural is a seasoned Machine Learning Engineer at Voxel51 with a strong passion for empowering Data Scientists and ML Engineers to unlock the full potential of their data.
Resource links
- Schedule a FiftyOne Teams demo to see the MedSAM-2 and NVIDIA VISTA-3D models in action
- MedSam-2 dataset on Hugging Face
- Segment Anything in a CT Scan with NVIDIA VISTA-3D
Exploring Instance Imbalance in Medical Semantic Segmentation
Current benchmarks in Medical Semantic Segmentation either leave out imbalanced datasets or focus on class imbalance. However, the nature of semantic segmentation shows that it is construed towards the segmentation of objects without differentiating multiple instances within a single class. This leads to the problem of instance imbalance in semantic segmentation. This is quite concerning in the case of medical image segmentation where the size of instances is principal. This talk will focus on a new evaluation metric and analysis of losses particularly to understand instance imbalance in semantic segmentation.
Speaker: Soumya Snigdha Kundu is a Ph.D. student at King’s College London. His work is focused on Trustworthy Machine Learning (TML) and its application to Neuro-Oncology.
Resource links
- Soumya’s publications
Join the AI, Machine Learning and Computer Vision Meetup!
The goal of the Meetups is to bring together communities of data scientists, machine learning engineers, and open source enthusiasts who want to share and expand their knowledge of AI and complementary technologies.
Join one of the 12 Meetup locations closest to your timezone.
- Athens
- Austin
- Bangalore
- Boston
- Chicago
- London
- New York
- Peninsula
- San Francisco
- Seattle
- Silicon Valley
- Toronto
What’s Next?
Up next on Sept 26, 2024 at 2:00 PM BST / 6:30 PM IST , we have two great speakers lined up!
- GPUs at Scale – Trials of a GPUaaS Provider– Mischa van Kesteren, Solutions Engineer at NexGen Cloud
- Scaling Industrial AI with FiftyOne – Daniel Gural, Machine Learning Engineer at Voxel51
Register for the Zoom here. You can find a complete schedule of upcoming Meetups on the Voxel51 Events page.
Get Involved!
There are a lot of ways to get involved in the Computer Vision Meetups. Reach out if you identify with any of these:
- You’d like to speak at an upcoming Meetup
- You have a physical meeting space in one of the Meetup locations and would like to make it available for a Meetup
- You’d like to co-organize a Meetup
- You’d like to co-sponsor a Meetup
Reach out to Meetup co-organizer Jimmy Guerrero on Meetup.com or ping me over LinkedIn to discuss how to get you plugged in.
—
These Meetups are sponsored by Voxel51, the company behind the open source FiftyOne computer vision toolset. FiftyOne enables data science teams to improve the performance of their computer vision models by helping them curate high quality datasets, evaluate models, find mistakes, visualize embeddings, and get to production faster. It’s easy to get started, in just a few minutes.
