A Multistage Transfer Learning Approach for Acute Lymphoblastic Leukemia Classification

Automated medical image analysis driven by artificial intelligence can revolutionize modern healthcare in producing swift and precise diagnostics. Due to doctors' varying breadths of training and expertise, traditional leukemia screening methods frequently involve considerable subjectivity. Using a 3-stage transfer learning approach and stacks of convolutional neural networks, we constructed an efficient pathway for automatic leukemia identification and classification through various phases. Experimental findings disclosed that our pipeline powered by InceptionResNetV2 architecture decisively affects the accuracy with 99.60% (normal vs. leukemia) and 94.67% (normal to L3). Moreover, it reduces error rates by 1.65% and 6.05%, respectively. A consistent result via the T-test confirms our proposed framework robustness with a significant positive difference of 4.71% over the standard transfer learning mechanism (p-value = 0.0001 & t = 0.85310). This research could aid and support oncologists in early yet reliable prognoses of acute lymphoblastic leukemia types.