Enhanced Multi-Scale Convolutional Neural Network with Attention Mechanism for Accurate and Efficient Automated Hematological Diagnostics from Blood Smear
Santosh Kumar B, CH Hussaian Basha, Jini Kumari M, Pradeep Kumar S, V. Sathya, K Rajaram · 2025
Hematological disorders require accurate and timely diagnosis for effective treatment and patient management. This research develops a new framework that uses deep learning to rapidly identify blood disorders from peripheral blood pictures. Our system uses an AECM and MSCNN combination to diagnose anemia, infections, and leukemia with greater accuracy. The MSCNN captures image details at multiple scales while the AECM detects meaningful areas in blood images which boosts detection results. Our system reached 98.5% total accuracy after validating its performance on public blood smear images. The detection method delivered 97.8% precision and 98.2% recall values during testing. Our diagnostic framework makes blood test results quicker and more affordable while reducing patient labs demands in healthcare environments that face resource constraints.