Cervical Cytology Image Analysis and Classification Using Graph-Based Techniques and Deep Learning
Brown Bag Lecture by Sudhir Sornapudi | 11/27/2018 11:00AM – 12PM | 7th Floor Conference Room, Bldg 38A
Join WebEx meeting: https://nih.webex.com/join/jzounih.gov (access code if needed: 627 285 780)
For audio, if joining by phone, please dial 1-650-479-3208 (US/Canada)
Cervical cancer is one of the most common cancers in women. In 2018 in the U.S., it is estimated that about 13,240 cases will be diagnosed for invasive cervical cancer with about 4,170 deaths. The use of Pap smear (liquid-based cytology) has significantly contributed to reducing mortality. Goals of this research are detection and classification of worst grade cells in whole slide images of liquid-based Pap cytology slides. We explore deep learning through fine tuning and transfer learning of CNN models pre-trained on natural images and fine-tuned using cervical cells from Herlev Pap smear dataset. The trained model is validated using cellular image patches extracted from 25 cytology whole slide image scans. We highlight the challenges in cytology image analysis and in generating ground truth from the microscopic slide data. We also describe detection of overlapped cells by extracting sub-graphs from graph based cervical cell representation.
Sudhir is a second year Ph.D. candidate in Computer Engineering at Missouri University of Science and Technology (MUST), Rolla. He is currently working as a Graduate Research Fellow at NLM with Sameer Antani and Rodney Long. He earned his Masters in Computer Engineering from MUST in 2017 and his Bachelors in Electronics and Communications Engineering from Jawaharlal Nehru Technological University, India in 2014. Sudhir’s expertise include image processing and deep learning applications in biomedical imaging. He has experience in working with datasets like cervical histology images, foot prosthetics and GI Tract.