Retrieval Augmented Generation for Pathology Reports
Tanya Nair, Connor Friedman & Aayush Shivashankar
Lay Summary:
Our project creates a tool to help clinicians gain a better understanding of their patients' cancer cases by summarizing key details of relevant cancer cases from a large database of pathology reports. This tool uses an efficient artificial intelligence model to provide relevant insights without requiring extensive computing resources or internet access.
Abstract:
Health literacy and clear communication are essential for informed decision-making and optimal care. With the current complex, unstandardized structure of pathology reports, many clinicians face challenges in interpreting these reports effectively and understanding their patients’ cases thoroughly. This study aims to create a tool for clinicians to advance their understanding of patient cases via use of a SLM-based RAG model that promotes greater understanding of patient cases for oncologists. We hypothesize that this approach will enhance both information extraction and classification of pathology reports in a more efficient manner than currently implemented methods. The model utilizes an information retrieval system with a fusion search method based on Okapi BM25 and FAISS quantized vector search with 9,523 pathology reports to draw from The Cancer Genome Atlas program. Running on just 6.9 gigabytes of RAM with no internet connection required, our model aims to enhance clinicians’ knowledge and awareness of different cancer variants by pulling relevant pathology reports from the database and using the SLM phi-2 to summarize and standardize them for clinicians’ learning. This system enhances understanding and communication in clinical settings with lower resource requirements compared to large language models. Future work will explore further customizations, integrations, and specializations to address specific cancer types, incorporate recent research findings, and adapt the model to various use cases based on location-specific data.
Q&A:
Bios: Tanya Nair,Connor Friedman,Aayush Shivashankar,Valmik Nahata
Program Track: Advanced Research
GitHub Username:
t-nair -Tanya Nair
ConnorFriedman10 -Connor Friedman
aay18-aar12 -Aayush Shivashankar
Valpip123EMY -Valmik Nahata
What was your favorite seminar? Why?
My favorite seminar was the Cervical Cancer Screening seminar by Onyi, it was very interesting to learn about her research and her experiences at Dartmouth as an undergrad. -Tanya Nair
By far my favorite seminar was the one presented by Onyinyechi Owo, which I found interesting not only because it directly addressed my groups topic (which allowed us to meet with her later and gain an even deeper understanding of our topic and overall goal), but also because the attitude it was presented with was extremely delightful and unlike nearly every other seminar. -Connor Friedman
My favorite seminar was with Onyinyechi Owo because her entire seminar regarding her work with SCC and other skin cancers was so close to our project goals. Furthermore, we were able to also reach out to her after and have a discussion with her allowing us to further our project by actually talking to someone in the field. -Aayush Shivashankar
My favorite seminar was with Onyinyechi Owo as her work and knowledge of skin cancers served as an inspiration for us. -Valmik Nahata
If you were to summarize your summer internship experience in one sentence, what would it be?
I’m incredibly appreciative of my experience with Levy Lab as a summer intern, I learned a lot about collaboration in a research setting and gained significant technical knowledge, both of which I know will serve me well in the future. -Tanya Nair
This internship has granted me an insight into the medical applications of machine learning unlike anything I’ve ever experienced, and throughout this program I’ve grown to not only be a better researcher, but also a better student. -Connor Friedman
A rollercoaster of information–but altogether something I genuinely enjoyed because of the new knowledge and friends I made along the way. -Aayush Shivashankar
I enjoyed worked with a team of students to develop an extensive research project from scratch. -Valmik Nahata