This repository contains clinical applications of AMAP (Automatic Morphological Analysis of Podocytes), a deep learning tool for segmenting and quantifying podocyte foot process morphology from high-resolution fluorescent microscopy images.
[LINK TO PAPER]
Podocyte injury is present in most glomerulonephrities and causes different subtypes of FSGS lesions which predict progression in IgA Nephropathy. This study applies high-resolution confocal microscopy with AMAP to identify novel morphometric fingerprints for progressive disease in IgAN and other forms of glomerular diseases.
[DESCRIPTION OF CLINICAL APPLICATION AND MODIFICATIONS]
The original AMAP model was fine-tuned on clinical biopsy data to improve segmentation accuracy on clinical samples. [MORE DETAILS ABOUT FINE-TUNING PROCEDURE WILL BE ADDED]
The clinical data used in this study cannot be publicly shared due to privacy regulations. However, the annotation procedure and fine-tuning approach will be explained in detail in the associated publication.
- amap - Original AMAP research code, modified for Python 3.11 compatibility
- amap-app - Desktop application version with:
- Hole-filling code for improved segmentation
- Fine-tuned model checkpoint
See individual README files in amap/ and amap-app/ for installation and usage instructions.