GAMF-Net: Grouped Availability-Aware Multimodal Fusion for Smartphone Skin Lesion Classification
To address the lack of structured organization among heterogeneous clinical variables, the ambiguity between observed and imputed values, and the risk that clinical information may interfere with visual evidence, we propose GAMF-Net, a grouped availability-aware multimodal fusion network.
ProofPaper ↗
Key points
- Smartphone clinical images provide a low-cost data source for computer-aided skin lesion classification in resource-constrained settings, but images alone cannot fully capture patients' medical histories and lesion symptoms.
- Ten clinical variables are divided into three semantic groups---demographics, history-related risk, and lesion presentation---and each group jointly encodes field values and field-level availability masks.
- We evaluate the method on PAD-UFES-20 under a patient-disjoint five-fold protocol with separate training, validation, and test partitions.
- GAMF-Net achieves an Accuracy of $74.17\pm2.85%$, a balanced accuracy (BACC) of $70.11\pm2.11%$, a Macro-F1 of $68.85\pm0.85%$, and a Macro-AUROC of $93.42\pm1.20%$, outperforming all image-only and multimodal baselines reimplemented under a unified protocol.
Sources (1)
- [1]GAMF-Net: Grouped Availability-Aware Multimodal Fusion for Smartphone Skin Lesion ClassificationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:35 PM
To address the lack of structured organization among heterogeneous clinical variables, the ambiguity between observed and imputed values, and the risk that clinical information may interfere with visual evidence, we propose GAMF-Net, a grouped availability-aware multimodal fusion network.
Smartphone clinical images provide a low-cost data source for computer-aided skin lesion classification in resource-constrained settings, but images alone cannot fully capture patients' medical histories and lesion symptoms.
Extractive summary: sentences quoted from the sources.