A Deep Learning Framework for Squamous Cell Segmentation in Whole Slide Images and Feature Extraction for Cervical Cancer Diagnostics
Maya Sarkar, Vivaan Mahajan, Lacie Zhou, Doniyor Babajanov, Neha Ranjith & Neha Shaik
Lay Summary:
We developed an AI system that can automatically find and analyze cervical cells in microscope images, helping doctors detect early signs of cervical cancer more quickly and accurately. This technology could make life-saving screening more accessible in places where trained specialists and advanced medical tools are limited.
Abstract:
Cervical cancer (CC) remains a leading cause of preventable death among women, particularly in low-resource settings where access to early diagnostic tools is limited. Existing screening techniques such as Pap smears, colposcopy, and HPV testing require trained professionals and are often inaccessible in rural regions. In this study, a deep learning approach was trained and evaluated on a dataset of 1,038 whole slide images (WSIs) to enhance CC screening for automated cytopathological analysis. Using the YOLOv8 model, squamous cells were precisely segmented through high-resolution segmentation masks, which were then processed by a foundation model to generate 512-dimensional feature embeddings (FEs) from the second to last layer. These embeddings served as rich descriptors for downstream diagnostic tasks like clustering squamous cells based on morphological features. With the help of GPU acceleration and batch processing, our methodology was optimized for efficiency in order to facilitate the high-throughput analysis necessary for large WSI datasets. As a result of this work, a significant step towards augmenting the capabilities of pathologists has been made, improving the efficiency of CC screening and the potential to support equitable CC screening worldwide. Keywords—cervical cancer, YOLO, machine learning, cell detection
Q&A:
Bios: Maya Sarkar,Vivaan Mahajan,LangXi (Lacie) Zhou,Doniyor Babajanov,[Neha Ranjith](https://ben-m-7.github.io/editai_internship/people/,Neha Shaik
Program Track: Advanced Research
GitHub Username:
mayasarkar -Maya Sarkar
Name: Vivaan Mahajan vivaanmahajan14 Vivaan Mahajan vivaanmahajan14 Name: GitHub Username:, dtype: object -Vivaan Mahajan
doniyor-babajanov -Doniyor Babajanov
A-Red-Hen -Neha Shaik
What was your favorite seminar? Why?
My favorite seminar was the one on conducting successful research because it gave useful tips on how to retain information, such as writing down detailed notes throughout the process of your project. -Maya Sarkar
Name: Vivaan Mahajan My favorite seminar was the one where Dr. Haze… Vivaan Mahajan My favorite seminar was the one where Dr. Haze… Name: What was your favorite seminar? Why?, dtype: object -Vivaan Mahajan
I enjoyed nearly every seminar, but (if I had to pick) I’d say the “Hallmarks of Precancer” seminar by Ken Lau was especially interesting as it approached cancer by going into how it originated and was especially relevant to our project. I also loved Lau’s easy-going personality and appreciated how he made a visible effort to relate the content to our summer projects. Thanks Dr. Lau!!! -Doniyor Babajanov
I liked Rishy and Vismay’s seminar on ALS since I knew a little bit about that project before and thought the direction they took with it was cool -Neha Shaik
If you were to summarize your summer internship experience in one sentence, what would it be?
My summer internship this year I had the amazing opportunity to consistently meet with mentors for guidance and work with my team members. -Maya Sarkar
Name: Vivaan Mahajan I learned how to conduct an independent resear… Vivaan Mahajan I learned how to conduct an independent resear… Name: If you were to summarize your summer internship experience in one sentence, what would it be?, dtype: object -Vivaan Mahajan
An intense learning opportunity which accelerated my growth in deep-learning and in collaborative, professional research. -Doniyor Babajanov
I loved my time at the EDIT ML Summer Internship because it provided me with the opportunity to work on projects with clinical significance and develop my skills under the mentorship of experienced researchers. -Neha Shaik