Blood Group Identification from Fingerprint Images Using Deep Learning and Fingerprint Sensor Integration

This study investigates an exploratory approach to classifying four positive blood-group labels— A+, B+, AB+, and O+—from fingerprint images using deep learning and a Mantra MFS100 fingerprint sensor. A canonical fingerprint dataset containing approximately 6000 images was used for model development. The selected four-class model reportedly achieved 87.89% accuracy (370/421) on the canonical test set. In a separate exploratory MFS100 evaluation, 78 images were grouped into 26 timestamp-defined capture bursts; the reported image-level accuracy was 46.15% (36/78), with strong prediction bias toward O+. Available metadata do not establish biological donor independence between groups. Three consecutive left-thumb scans are aggregated by averaging class probabilities, although an accuracy improvement from this strategy has not been established. The system is a research prototype and is not a substitute for clinical blood typing.