Digital Pathology Image Analysis With Reusable Container Files

The patent applies preliminary action by pre-processing digital pathology images into standardized container files with embedded metadata (including magnification level, color channel, and annotation information) before training begins. This pre-processing step is performed once and reused across multiple training iterations, eliminating the need to reprocess raw images during each training cycle. The container files are structured with rasterized tiles and associated metadata that can be directly loaded by ML models, saving significant computational time while maintaining access to large datasets for improved model performance.

The patent extracts and stores essential metadata (magnification level, color channel, annotations) alongside the image data in container files. This extraction allows the ML training process to access only the necessary information from large datasets without processing unnecessary data, reducing computing resource demands. The metadata enables efficient filtering and selection of relevant training samples, allowing models to be trained on large datasets with reduced computational overhead.

The patent performs preliminary preprocessing of digital pathology images into standardized container files that embed metadata (magnification level, color channel, annotations) and organize data into reusable formats. This one-time preprocessing effort enables the same processed data to be reused across multiple ML training projects without reprocessing raw images. The standardized container structure facilitates easy sharing and distribution of training datasets while maintaining data integrity and metadata associations.

The patent merges the image data (rasterized tiles) with their associated metadata (magnification level, color channel, annotations) into single container files. This combining eliminates the need for separate storage of images and metadata, reducing overall storage requirements while maintaining all necessary information for ML training. The unified container format simplifies data sharing and distribution, as complete training datasets can be transferred as self-contained files without requiring separate metadata files or additional processing upon receipt.