In India, Anemia continues to be a major public health concern, but current diagnostic factors are often based only on individual clinical or demographic markers and leave out the multidimensional variables that determine the situation. This research proposes an AnemiaFusionNet transformer-based framework that employs Text based clinical features, conjunctival image measure, and geospatial data to optimize performance of anemia detection. Clinical or textual data and geospatial information were taken from the National Family Health Survey (NFHS), whereas imagery data composed of piloted palpebral conjunctival images. The conjunctival image set was examined through convolutional encoders and reviewed alongside NFHS-based clinical and locational specifications to enable rigorous multimodal modeling. Firstly, we determined the individual modality models to acquire baseline performance, next we fused the heterogeneous embeddings using a cross-attention transformer section. The fused model constantly surpassed outcomes from unimodal techniques, with performance improvements and improved generalization to subgroups of the population. Spatial representations derived from the integrated model shows persistent high-risk areas across central and eastern part of India. The results underscore the complementary predictive value of visual, clinical, and locational cues, and demonstrate that multimodal learning provides a more detailed and operationally effective approach for early anemia detection and geographically intended public health intervention planning.
Keywords: NFHS Data, AnemiaFusionNet, Transformer Architecture, Deep Learning, Cross-Modal Fusion, Machine Learning
How to cite this article: Minocha N, Yagnik S, Short Title: Anemia Fusion Net for Multimodal Anemia Detection Anemia Fusion Net: A Next-Generation Transformer Architecture for Anemia Detection. Int J Drug Deliv Technol. 2026;16(4s): 9-16; DOI: 10.25258/ijddt.16.9-16