Toward Driverless AI: Automating Leukemia Detection and Classification using Hyperautomation, a Case Study

Abstract

This paper mainly explores how driverless AI can help researchers accurately classify leukemia diagnosis. Since Leukemia is among the most common commonly occurring types of cancer and thus early detection and classification can be of paramount significance. In this paper, we proposed a reliable method to automate the classification of leukemia in which we used the automated machine learning (henceforth, AutoML) library available in H2O.ai to select the best classification model capable of detection and classification depending on peripheral blood smear (PBS) images. The proposed framework used three datasets of bone marrow images to detect Leukemia and accordingly to classify blood cells. we compared some previous studies conducted in this field adopting traditional classification approaches with our proposed model of hyperautomation. We divide these data between training and testing using threefold cross-validation. The final AI model produced the most accurate diagnostics, and its effectiveness was evaluated using the accuracy, specificity, F-score, recall, precision, and Area Under the Curve (AUC) from the Receiver Operating Characteristic (ROC) curve, and it achieves an elegant performance for Leukemia classification with AUC of 1 and 1 in the testing and validation sets, accordingly. Results indicate that almost every model trained on H2O.ai driverless works relatively better than other models that work with traditional software with less time and effort.