Classification of renal biopsy direct immunofluorescence image using multiple attention convolutional neural network

Area of Science:

• Nephrology
• Medical Imaging
• Artificial Intelligence

Background:

• Direct immunofluorescence (DIF) is crucial for renal pathology, aiding glomerulonephritis diagnosis by analyzing immunoglobulin deposition patterns.
• Manual classification of these patterns is labor-intensive and prone to inter-observer variability.
• Automating the identification and fusion of deposition location and appearance can enhance diagnostic efficiency.

Purpose of the Study:

• To develop an automated framework for segmenting glomeruli and classifying immunoglobulin deposition patterns in DIF images.
• To fuse classification results for deposition location and appearance to assist physicians in generating immunofluorescence reports.

Main Methods:

• A framework combining a pre-segmentation module and a classification module was proposed.
• A segmentation network identified glomeruli, followed by a multiple attentions convolutional neural network (MANet) for classifying deposition region and appearance.
• Classification results from two pre-trained networks were fused with labels.

Main Results:

• The framework achieved high classification accuracy: 98% for deposition region and 95% for deposition appearance.
• Accurate fusion of deposition appearance and classification labels was demonstrated.
• The system effectively automates patterned immunofluorescence report generation.

Conclusions:

• Automated generation of patterned immunofluorescence reports is achievable with high accuracy.
• This AI-driven approach can significantly improve the diagnosis of autoimmune kidney diseases.
• The proposed method offers a valuable tool for enhancing clinical decision-making in nephrology.