Segmentation algorithm can be used for detecting hepatic fibrosis in SD rat.

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Liver fibrosis, an early stage of liver cirrhosis, is a critical target for antifibrotic drug discovery.
• Current antifibrotic agents often fail in clinical trials due to adverse reactions, necessitating robust non-clinical evaluation methods.
• Automated quantification of hepatic fibrosis using artificial intelligence (AI) is emerging, but optimal algorithm performance remains unevaluated.

Purpose of the Study:

• To evaluate and compare the performance of three deep learning localization algorithms (mask R-CNN, DeepLabV3+, and SSD) for detecting hepatic fibrosis in non-clinical studies.
• To identify the most accurate AI algorithm for quantifying hepatic fibrosis in rodent models.

Main Methods:

• Trained three deep learning algorithms—mask R-CNN, DeepLabV3+, and SSD—on 5750 images with 7503 annotations of hepatic fibrosis.
• Evaluated model performance using precision and recall metrics on large-scale images.
• Compared algorithm predictions against manual annotations for accuracy.

Main Results:

• Mask R-CNN achieved the highest recall (0.93), demonstrating the closest predictions to manual annotations for hepatic fibrosis detection.
• DeepLabV3+ showed good performance but misclassified hepatic fibrosis as inflammatory cells and connective tissue.
• SSD exhibited the lowest performance with a recall of 0.75, indicating limitations in predicting hepatic fibrosis.

Conclusions:

• Segmentation-based AI algorithms, particularly mask R-CNN, show significant promise for accurate hepatic fibrosis prediction in non-clinical research.
• AI-driven image analysis can enhance the evaluation of antifibrotic agents in preclinical settings.
• Further application of AI segmentation algorithms is recommended for reliable non-clinical hepatic fibrosis assessment.