Multimodal Knowledge Aggregation for Microbial Keratitis: Geographic Confounding and Shortcut Risk in Pooled Multicountry Data
Abstract
To develop an open-source multimodal model for bacterial-versus-fungal microbial keratitis classification and assess shortcuts from geographic pooling. Participants: 1,645 US and Indian patients contributed 17,158 slit-lamp images under blue-light, sclerotic-scatter, and white-light illumination. Methods: The model combined cross-modality contrastive learning, modality-specific fine-tuning, and fusion with 7 clinical variables. Diagnoses used smear, culture, and polymerase chain reaction results, with expert review of culture-negative cases. The model and 5 comparators used 5 patient-disjoint outer folds stratified by country and organism, with inner validation. Secondary analyses examined within-country performance, a country-only reference, ablations, and imbalance mitigation. Code: https://github.com/yqwang01/TPMKA. Main Outcome Measures: Patient-level macro-F1, class-specific F1, accuracy, and area under the receiver operating characteristic curve (AUROC), overall and by country. Results: The model had the highest overall macro-F1 (83.06%; 95% CI, 81.58-84.56) and accuracy (84.50%; 95% CI, 83.16-85.84) among matched comparators, with 89.9% AUROC (95% CI, 88.7-91.1). Paired differences were not consistently significant. AUROC was 83.9% in India and 67.0% in the US; bacterial F1 in India was 47.78%, and fungal F1 in the US was 40.97%. A country-only reference achieved 81.88% accuracy and 78.2% AUROC overall despite 50.0% AUROC within each country. Resampling and group distributionally robust optimization improved some underrepresented groups without removing the geographic pattern. Conclusions: The model compared favorably with matched methods, but geographic prevalence differences made pooled performance more reassuring than within-country discrimination. Independent geographic validation is needed.
Related
- A Test for Confounding in Coupling of Multimodal Neuroimaging Data
- MINT: Predicting Enzyme-Cofactor Specificity by Multimodal Knowledge Transfer from Protein-Ligand Interactions
- Metabolite-centric Knowledge Graph Unveils Predictive and Mechanistic Insights through Self-supervised Multimodal Learning
- Feasibility and immediate effects of a multimodal, dance-integrated educational video package on eye health knowledge and intentions in Nigeria: A pre-post pilot study
- Diagnostic frameworks and abortifacient findings in domestic ruminant abortions: a multicountry European study (2022–2024)
BibTeX
@misc{wang2026multimodal,
title = {{Multimodal Knowledge Aggregation for Microbial Keratitis: Geographic Confounding and Shortcut Risk in Pooled Multicountry Data}},
author = {Yiqing Wang and Maria A. Woodward and Ziyun Yang and N. Venkatesh Prajna and Chunming He and Leslie M. Niziol and Mercy Pawar and Ming-Chen Lu and Guillermo Amescua and Rachel A. F. Wozniak and Sejal Amin and Abinaya Krishnan and Prabhleen Kochar and Sina Farsiu},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2607.03740},
eprint = {2607.03740},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2607.03740},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Multimodal Knowledge Aggregation for Microbial Keratitis: Geographic Confounding and Shortcut Risk in Pooled Multicountry Data
AU - Yiqing Wang
AU - Maria A. Woodward
AU - Ziyun Yang
AU - N. Venkatesh Prajna
AU - Chunming He
AU - Leslie M. Niziol
AU - Mercy Pawar
AU - Ming-Chen Lu
AU - Guillermo Amescua
AU - Rachel A. F. Wozniak
AU - Sejal Amin
AU - Abinaya Krishnan
AU - Prabhleen Kochar
AU - Sina Farsiu
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2607.03740
UR - https://arxiv.org/abs/2607.03740
AB - To develop an open-source multimodal model for bacterial-versus-fungal microbial keratitis classification and assess shortcuts from geographic pooling. Participants: 1,645 US and Indian patients contributed 17,158 slit-lamp images under blue-light, sclerotic-scatter, and white-light illumination. Methods: The model combined cross-modality contrastive learning, modality-specific fine-tuning, and fusion with 7 clinical variables. Diagnoses used smear, culture, and polymerase chain reaction results, with expert review of culture-negative cases. The model and 5 comparators used 5 patient-disjoint outer folds stratified by country and organism, with inner validation. Secondary analyses examined within-country performance, a country-only reference, ablations, and imbalance mitigation. Code: https://github.com/yqwang01/TPMKA. Main Outcome Measures: Patient-level macro-F1, class-specific F1, accuracy, and area under the receiver operating characteristic curve (AUROC), overall and by country. Results: The model had the highest overall macro-F1 (83.06%; 95% CI, 81.58-84.56) and accuracy (84.50%; 95% CI, 83.16-85.84) among matched comparators, with 89.9% AUROC (95% CI, 88.7-91.1). Paired differences were not consistently significant. AUROC was 83.9% in India and 67.0% in the US; bacterial F1 in India was 47.78%, and fungal F1 in the US was 40.97%. A country-only reference achieved 81.88% accuracy and 78.2% AUROC overall despite 50.0% AUROC within each country. Resampling and group distributionally robust optimization improved some underrepresented groups without removing the geographic pattern. Conclusions: The model compared favorably with matched methods, but geographic prevalence differences made pooled performance more reassuring than within-country discrimination. Independent geographic validation is needed.
ER -