Implementing deep learning models for the classification of Echinococcus multilocularis infection in human liver...

Area of Science:

• Pathology
• Medical Imaging
• Computational Biology

Background:

• Histological diagnosis of alveolar echinococcosis is challenging.
• Deep learning (DL) aids pathologists, but data on parasitic infections are scarce.
• Need for DL models to classify Echinococcus multilocularis liver lesions.

Purpose of the Study:

• Implement DL methods for classifying E. multilocularis liver lesions and normal liver tissue.
• Assess critical regions and structures for classification decisions.
• Enhance diagnostic support for pathologists.

Main Methods:

• Extracted 15,756 echinococcus tiles and 11,602 normal liver tiles from whole slide images (WSI).
• Utilized pretrained DL model architectures with 60-20-20% data splitting.
• Visualized predictions with heat maps and employed GradCAM for spatial feature analysis.

Main Results:

• DL models achieved high validation and test set accuracy with an Area-Under-the-Curve (AUC) of 1.0.
• GradCAM identified pericystic fibrosis, necrotic areas, and metacestode layers as crucial for classification.
• Models demonstrated excellent predictive performance.

Conclusions:

• Deep learning models show high predictive performance for classifying E. multilocularis liver lesions.
• Future work includes validating models with diverse datasets and testing against other parasitic entities like Echinococcus granulosus.