Maximizing the Accuracy of a Thinprep Cell Analysis Model on Surepath
Amrit Singh & Shreyas Pendem
Lay Summary:
We're improving a computer model that helps doctors detect cervical cancer by making it work better with different testing methods. This will help ensure that more women get accurate and reliable results, no matter which test is used.
Abstract:
Cervical cancer and cervical intraepithelial neoplasia (CIN) are significant health concerns affecting millions of women globally. Liquid-based cytology (LBC), a key diagnostic method, utilizes two primary platforms- Thinprep and Surepath, both essential for detecting cervical abnormalities. Dartmouth’s current model effectively classifies benign and malignant cells in Thinprep samples, but it is not optimized for Surepath, limiting its applicability across different cytology platforms. This project seeks to adapt the Dartmouth model to accurately analyze Surepath samples by employing CycleGAN architecture, known for its proficiency in image-to-image translation. By retraining the model, we aim to enhance its performance and generalization across Surepath samples, addressing a crucial challenge in ensuring consistent diagnostic outcomes. The study examines whether transferring learning from the Thinprep-optimized model to Surepath is more effective than retraining from scratch, and if transfer learning proves beneficial, it will explore the most efficient strategies for its implementation. This research has the potential to significantly improve diagnostic accuracy in LBC, leading to better patient outcomes by ensuring that the model generalizes effectively across different cytology methods, contributing to the broader field of medical image analysis and addressing the challenges of cross-domain adaptation in digital pathology.
Q&A:
Bios: Amrit Singh,Shreyas Pendem
Program Track: Advanced Research
GitHub Username:
AmritSingh10 -Amrit Singh
shreyas12074472 -Shreyas Pendem
What was your favorite seminar? Why?
Jason Wei’s seminar. This is because I am very interested in OpenAI, especially Generative AI, and I think that he offered valuable insights into them. I am also very interested in going down a similar path that he did, so I found his seminar very fascinating. -Amrit Singh
My favorite seminar was on using multiple sources of information to make a diagnosis for a patient. Though I do not remember the individual who presented this information, he made it clear that patients are diagnosed considering Whole Slide Images (WSI), Spatial Transcriptomics (ST), and other information. The idea was that machine learning models should be able to do this too. This presentation illustrated how a simple idea came out effective. I liked how the presenter shared his thoughts and how he gave some inspiration for high schoolers who are completing research. -Shreyas Pendem
If you were to summarize your summer internship experience in one sentence, what would it be?
I learned a lot on topics that I didn’t even knew existed, and learned the amount of work that goes into research and pathology. -Amrit Singh
A valuable learning hands-on experience of research in pathology with the assistance of professors and mentors. -Shreyas Pendem