Creating an atlas of normal tissue for pruning WSI patching through anomaly detection

Area of Science:

• Computational pathology
• Digital pathology
• Histopathology image analysis

Background:

• Whole slide images (WSIs) analysis often includes normal histology, which can be redundant.
• Existing computational pathology methods overlook the impact of normal tissue on analysis.
• Efficiently identifying and excluding normal tissue is crucial for accurate pathological lesion detection.

Purpose of the Study:

• To propose and validate a novel method for creating a "normal tissue atlas" using only normal biopsy WSIs.
• To demonstrate how this atlas can enhance the representativeness of selected WSI patches by eliminating normal tissue fragments.
• To improve the performance of search engines and analytical tools in computational pathology.

Main Methods:

• Construction of a normal tissue atlas using 107 normal skin WSIs.
• Application of the atlas to filter patches from WSIs of cutaneous squamous cell carcinoma (553 WSIs) and breast cancer (451 WSIs).
• Evaluation of the impact on patch selection and search performance using leave-one-patient-out validation.

Main Results:

• The normal atlas reduced the number of selected WSI patches by 30% to 50%.
• Indexing and search performance were maintained after utilizing the normal atlas.
• Validation on both skin and breast cancer datasets confirmed the method's effectiveness.

Conclusions:

• The proposed normal tissue atlas is a promising approach for unsupervised selection of representative abnormal WSI patches.
• This method enhances the efficiency and accuracy of computational pathology by focusing on diagnostically relevant tissue.
• The atlas facilitates improved analysis by reducing data volume while preserving essential pathological information.