MLCD: A Unified Software Package for Cancer Diagnosis

Area of Science:

• Digital pathology
• Machine learning in oncology
• Computational cancer diagnosis

Background:

• Breast cancer diagnosis relies on histopathological analysis of biopsy slides.
• Improving diagnostic accuracy and efficiency is crucial for patient outcomes and research.
• Existing methods can be subjective and time-consuming.

Purpose of the Study:

• To develop a unified software package (MLCD) integrating machine learning algorithms for breast cancer biopsy diagnosis.
• To enhance the quality of clinical practice and cancer research through improved diagnostic tools.
• To provide open-source, accessible tools for researchers and clinicians.

Main Methods:

• Utilized whole-slide images from 240 breast biopsy cases for algorithm development and model training.
• Developed methodology for identifying regions of interest (ROIs) in whole-slide images.
• Implemented semantic segmentation for tissue classification and diagnostic classification into four categories (benign, atypia, ductal carcinoma in situ, invasive cancer).

Main Results:

• The MLCD package includes an ROI detector and modules for semantic segmentation and diagnostic classification.
• The software classifies ROIs into benign, atypia, ductal carcinoma in situ, and invasive cancer categories.
• The package is available via GitHub under an MIT license, with future integration into the Pathology Image Informatics Platform.

Conclusions:

• The developed tools have the potential to assist cancer researchers and practicing physicians.
• This software aims to motivate future research in computational pathology and cancer diagnosis.
• The article details the software's development methodology and provides sample outputs for user guidance.