MACN: A Modality-Aware Cross-Attention Transformer for Non-Invasive Hemoglobin Estimation and Anemia Classification Using Multi-Region Physiological Images

Anemia is a major health concern among pregnant women in rural areas. Early detection and timely treatment are necessary to reduce maternal and infant mortality. Many non-invasive techniques have been proposed to solve the problem of unavailability of laboratory-based hemoglobin testing in rural areas. But their accuracy is limited due to dependence on a single modality. This work proposes a multimodal cross-attention network (MCAN) for anemia classification and hemoglobin (Hgb) level estimation using multiple physiological visual cues. Deep convolutional features extracted from four regions of the eye conjunctiva, fingertip, palm, and lip mucosa are fused using a cross-attention multimodal fusion. The fused features are used to estimate Hgb level and predict anemia. Through experimental analysis, the proposed solution is found to increase the estimation accuracy by 5% compared to existing single approaches. The statistical significance of the performance gain is validated through bootstrap confidence intervals and DeLong’s test.

@article{P2026,
  title     = {MACN: A Modality-Aware Cross-Attention Transformer for Non-Invasive Hemoglobin Estimation and Anemia Classification Using Multi-Region Physiological Images},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Chaitra S P and D R Ramesh Babu},
  doi       = {10.14569/IJACSA.2026.0170763},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170763}
}