GLAM: Glomeruli Segmentation for Human Pathological Lesions using Adapted Mouse Model.

Area of Science:

• Nephrology and computational pathology
• Biomedical image analysis
• Translational research

Background:

• Accurate kidney tissue measurement is crucial for drug development and understanding diseases.
• Existing glomeruli segmentation techniques show promise for mouse-to-human translation but often overlook complex pathological lesions.
• Pathological glomeruli are more clinically relevant and data from animal models can be readily scaled.

Purpose of the Study:

• To investigate the effectiveness of applying pathological segmentation models trained on mouse models to human patients.
• To introduce GLAM, a deep learning approach for fine-grained segmentation of human kidney lesions using mouse models.
• To evaluate zero-shot and hybrid learning strategies for mouse-to-human transfer learning in kidney lesion segmentation.

Main Methods:

• Developed GLAM, a deep learning framework for segmenting human kidney lesions.
• Employed mouse models for training segmentation algorithms.
• Evaluated zero-shot transfer learning and hybrid learning strategies, leveraging mouse samples for human data segmentation.

Main Results:

• Hybrid learning demonstrated superior performance in segmenting human pathological kidney lesions compared to other strategies.
• The study successfully addressed the challenge of mouse-to-human transfer learning for complex lesion segmentation.
• GLAM framework facilitated fine-grained segmentation of human kidney lesions.

Conclusions:

• Hybrid learning is a highly effective strategy for transferring kidney lesion segmentation models from mice to humans.
• The findings support the clinical utility of deep learning models trained on animal data for human pathological analysis.
• This research advances the application of AI in nephrology for improved diagnostics and treatment development.