Case Studies

The Chinese University of Hong Kong deployed DCAN (Deep Contour-Aware Networks), a deep convolutional neural network for automated gland segmentation in colorectal cancer histology images. The system won the 2015 MICCAI Gland Segmentation Challenge, achieving an F1 score of 0.912 on benign tissue and 0.716 on malignant tissue, while training 100 times faster on NVIDIA TITAN X GPUs compared to CPUs.

NYU Langone Health deployed a deep learning system that matches the accuracy of board-certified radiologists in predicting breast cancer from MRI scans. The AI model, trained on over 21,000 MRI studies, achieved an AUROC of 0.924 on internal tests and demonstrated the potential to reduce unnecessary biopsies by up to 20% in BI-RADS 4 patients—those with suspicious findings requiring biopsy.