Deep Learning-Based Diagnostic Evaluation of Cholangiocarcinoma in Bile Duct Cytology Whole Slide Images
Vedhsai Thiriveedi & Aadi Duggal
Lay Summary:
We built an AI system that can look at bile duct cancer test slides under the microscope and help doctors tell apart healthy and cancerous cells more accurately. By combining smart image analysis with deep learning, our approach could make earlier and more reliable diagnoses for this deadly cancer.
Abstract:
Cholangiocarcinoma, or bile duct cancer, is a rare but highly lethal malignancy that is typically diagnosed at advanced stages due to subtle and nonspecific early symptoms. This study investigates the integration of Whole Slide Imaging (WSI) with advanced deep learning methods to improve diagnostic performance in bile duct brushing cytology. WSIs were annotated using the Segment Anything Model (SAM) and analyzed with YOLO-based object detection to localize cell clusters. Cluster-level images were subsequently classified with convolutional neural networks (CNNs) and ResNet architectures. The models demonstrated strong performance- a binary AlexNet CNN achieved an area under the curve (AUC) of 0.94 and an accuracy of 88 percent, while ResNet-based approaches yielded accuracies exceeding 90 percent with robust generalization in cross-validation. These findings validate the feasibility of automated bile duct cytology interpretation and highlight the potential of deep learning models to substantially improve early detection and clinical decision-making. Future directions include developing models that classify real clusters into categories such as atypical, suspicious, malignant, and benign, performing connected component analysis (CCA) for automated segmentation, and creating three-dimensional or four-dimensional visualizations of subtype distributions. Together, these advances position deep learning as a transformative tool in Cholangiocarcinoma diagnostics.
Q&A:
Bios: Vedhsai Thiriveedi,Aadi Duggal
Program Track: Advanced Research
GitHub Username:
Name: Vedhsai Thiriveedi Vedhsai-codes Vedhsai Thiriveedi Vedhsai-codes Name: GitHub Username:, dtype: object -Vedhsai Thiriveedi
aadiduggal -Aadi Duggal
What was your favorite seminar? Why?
Name: Vedhsai Thiriveedi My favorite seminar was the one where we learn… Vedhsai Thiriveedi My favorite seminar was the one where we learn… Name: What was your favorite seminar? Why?, dtype: object -Vedhsai Thiriveedi
My favorite seminar was when Zarif Azher gave his presenation and he talked about him being a edit researcher for many years and now he is at Cal Tech! It was nice hearing from a student and someone from one of my dream colleges!! I enjoyed his presentation on applying ML to the medical field and him continuing a lot of his research in college especially. -Aadi Duggal
If you were to summarize your summer internship experience in one sentence, what would it be?
Name: Vedhsai Thiriveedi This summer, I developed and tested artificial… Vedhsai Thiriveedi My summer internship focused on developing and… Name: If you were to summarize your summer internship experience in one sentence, what would it be?, dtype: object -Vedhsai Thiriveedi
A challenging yet insightful and fun way to continue my research. -Aadi Duggal