HistoClean

HistoClean performs histological image pre-processing and augmentation to reduce bias and overfitting and improve the robustness of convolutional neural networks (CNNs) applied to digital pathology.

Key Features:

• Image Pre-processing and Augmentation: Performs comprehensive pre-processing and augmentation of histological images to increase consistency and variability for CNN training.
• Bias and Overfitting Mitigation: Applies augmentation strategies aimed at mitigating bias and reducing overfitting in deep learning models.
• Multi-level Accuracy Enhancement: Demonstrates improvements in model accuracy at tile, region of interest (ROI), and patient levels.
• Application to Stromal Maturity Detection: Supports development of models used for detecting stromal maturity from histological images.
• Support for CNN Development: Facilitates preparation of image data specifically for convolutional neural network workflows in image analysis.

Scientific Applications:

• Digital Pathology: Pre-processing and augmentation for deep learning workflows in digital pathology image analysis.
• Cancer Research: Improves predictive and prognostic CNN models used in cancer research.
• Stromal Maturity Assessment: Enables development of models to assess stromal maturity from histological sections.

Methodology:

Performs computational image pre-processing and augmentation of histological images for convolutional neural network development, with evaluation at tile, region of interest (ROI), and patient levels.