Multi-Modal Graph Neural Networks for Colposcopy Data Classification and Visualization DOI Open Access
Priyadarshini Chatterjee, Shadab Siddiqui, Priyadarshini Chatterjee

et al.

Cancers, Journal Year: 2025, Volume and Issue: 17(9), P. 1521 - 1521

Published: April 30, 2025

Cervical lesion classification is essential for early detection of cervical cancer. While deep learning methods have shown promise, most rely on single-modal data or require extensive manual annotations. This study proposes a novel Graph Neural Network (GNN)-based framework that integrates colposcopy images, segmentation masks, and graph representations improved classification. We developed fully connected graph-based architecture using GCNConv layers with global mean pooling optimized it via grid search. A five-fold cross-validation protocol was employed to evaluate performance before (1-100 epochs) after fine-tuning (101-151 epochs). Performance metrics included macro-average F1-score validation accuracy. Visualizations were used model interpretability. The achieved 89.4% accuracy 92.1% fine-tuning, which 94.56% 98.98%, respectively, fine-tuning. LIME-based visual explanations validated models focus discriminative regions. highlights the potential multi-modal analysis. Collaborating MNJ Institute Oncology, shows promise clinical use.

Language: Английский

Multi-Modal Graph Neural Networks for Colposcopy Data Classification and Visualization DOI Open Access
Priyadarshini Chatterjee, Shadab Siddiqui, Priyadarshini Chatterjee

et al.

Cancers, Journal Year: 2025, Volume and Issue: 17(9), P. 1521 - 1521

Published: April 30, 2025

Cervical lesion classification is essential for early detection of cervical cancer. While deep learning methods have shown promise, most rely on single-modal data or require extensive manual annotations. This study proposes a novel Graph Neural Network (GNN)-based framework that integrates colposcopy images, segmentation masks, and graph representations improved classification. We developed fully connected graph-based architecture using GCNConv layers with global mean pooling optimized it via grid search. A five-fold cross-validation protocol was employed to evaluate performance before (1-100 epochs) after fine-tuning (101-151 epochs). Performance metrics included macro-average F1-score validation accuracy. Visualizations were used model interpretability. The achieved 89.4% accuracy 92.1% fine-tuning, which 94.56% 98.98%, respectively, fine-tuning. LIME-based visual explanations validated models focus discriminative regions. highlights the potential multi-modal analysis. Collaborating MNJ Institute Oncology, shows promise clinical use.

Language: Английский

Citations

0