Multi-lens Neural Machine (MLNM 2.0.0)

This repository is the official implementation of the paper: An AI-assisted Tool For Efficient Prostate Cancer Diagnosis in Low-grade and Low-volume Cases.

It uses the data: Digital Pathology Dataset for Prostate Cancer Diagnosis.

We developed a multi-lens (or multi-resolution) deep learning pipeline detecting malignant glands in core needle biopsies of low-grade prostate cancer to assist pathologists in diagnosis. The pipeline consisted of two stages: the gland segmentation model detected the glands within the sections and the multi-lens model classified each detected gland into benign vs. malignant. The multi-lens model exploited both morphology information (of nuclei and glands from high resolution images - 40× and 20×) and neighborhood information (for architectural patterns from low resolution images - 10× and 5×), important in prostate gland classification.

In this release:

• class and function documentations were improved.
• patch cropping from Whole Slide Images were included.
• training and testing of a three-resolution benign vs. malignant classification model on the publicly available development set of the PANDA challenge were included.