Google AI Accelerator

DLBCL Cell Segmentation - AI-Powered Diagnosis with Mask R-CNN

Overview

As part of the Google AI Accelerator 2025 - a 2-week hands-on program for secondary and JC students in Singapore by SG Code Campus, IMDA, Google Cloud, and the Cancer Science Institute (CSI) at the National University of Singapore (NUS) and Dunman Secondary School. We developed an AI model to assist in diagnosing Diffuse Large B-Cell Lymphoma (DLBCL), the most common subtype of non-Hodgkin lymphoma. Accurate diagnosis of DLBCL requires labor-intensive examination of cell morphology by pathologists. Our solution uses AI to make this process faster and more consistent.

What I Used

- Languages/Tools: Python, Google Colab - Libraries/Models: Mask R-CNN, OpenCV, NumPy, PyTorch - Concepts: Instance segmentation, computer vision, medical imaging, BCL6 marker detection

My Role & Contributions

I worked on implementing and training the Mask R-CNN model to segment and classify BCL6-expressing cells in biopsy images. I also helped preprocess histopathological images, annotate training data, and evaluate segmentation accuracy using metrics like IoU and precision.

Key Features

• Mask R-CNN model trained on biopsy images for instance segmentation

• Highlights and classifies cells expressing B-cell lymphoma 6 (BCL6)

• Aims to assist pathologists by reducing time and improving accuracy

• Pipeline includes preprocessing, cell detection, classification, and visualisation

Predicted image generated by model

What I Learned

This project gave me hands-on experience with medical AI applications and advanced computer vision. I learned how to apply deep learning techniques like Mask R-CNN to real biomedical data, and gained insight into the intersection of AI and healthcare innovation.

Not found