Advancing the Diagnosis of Acute Lymphoblastic Leukemia with Hybrid Neural Networks

Acute Lymphoblastic Leukemia (ALL) is a malignancy of B or T lymphoblasts characterized by the uncontrolled replication of abnormal cells in the blood. The difficulty in diagnosing ALL arises from its visual similarity with other cells (HEM) in the blood, and misdiagnosis rates are high with other diseases. Thus, we seek an alternate solution to help medical professionals with the diagnosis of ALL. Recent solutions have utilized hybrid models that have delivered better performance than prior models, so we utilize a hybrid architecture unused for ALL detection. Our proposed architecture consists of EfficientNetB0 as our feature extractor and XGBoost classifier for its gradient boosting framework to perform a binary classification between ALL and HEM cells. The model maintains an accuracy of 85% when diagnosing ALL, proving the effectiveness of a data-driven approach for diagnosis. In the future, our model could be trained with other ALL datasets in order to become more versatile. Our work can assist doctors in providing an accurate and efficient diagnosis of leukemia, allowing for early intervention of the disease.