Machine Learning-Based Framework for Early Detection of Acute Leukemia to Reduce Disability Risk Through Timely Diagnosis

Immediate diagnosis of acute leukemia (AL) is essential for timely intrusion and successful patient outcomes, particularly for vulnerable groups as individuals with disabilities. This study proposes a framework to automate the speedy recognition of AL using cell population data (CPD), enabling the detection of abnormal samples with high precision while minimizing false positives. Further, we hypothesized early detection of hematological emergencies, particularly AL, in the pre-infinitesimal stage through related measurements of CPD. The predictive models were developed, trained, and validated using the NIBD dataset of 583 study subjects (including 361 AL cases). The proposed framework was designed to increase the accuracy of AL detection by flagging abnormal white blood cell counts. We identified specific CPD parameters, such as morphological factors, that effectively flag the presence of immature or abnormal white blood cells. Additionally, it differentiates between three groups—control, Acute Myeloid Leukemia (AML), and Acute Lymphoid Leukemia (ALL)achieving high area under the curve (AUC) values. The MLP classifier achieved the overall accuracy of 85%, F1 score of 91%, 81%, and 76% for control, AML, ALL respectively. This framework highlights the diagnostic speed and accuracy for AL patients and the importance of developing a system capable of detecting hematological emergencies.