Explainable AI For Leukemia Diagnosis: Interpreting Optimized Deep Learning Models for Improved Clinical Decision-Making

White blood cell morphology is the most important diagnostic component in the case of leukaemia. The need for a computerized system is appropriate in all cases since the judgment is impacted by the diagnosing the individual’s pathologic experiences. Under a microscope, the morphology of blood smear slides is used to make the diagnostic. An accurate diagnosis may be achieved with the use of deep and machine learning frameworks. Although these frameworks’ accuracy levels are excellent on their own, the outcomes of the diagnosis are much better when they are combined. The field of intelligent computing and machine learning has become more and more interested in ensemble learning strategies in recent decades. The use of already trained models during model construction is also beneficial for some designated tasks related to classification. All sorts of classifications make up this massive dataset that these algorithms are trained on. As a result, these models will work for a wide variety of circumstances. Transfer learning is used to tailor pre-trained models to the given challenge. Just as important for diagnosis is the ability to distinguish between subtypes of leukaemia using actual photos sourced from a privately held database, as opposed to relying only on binary classification on pictures taken from standard datasets. The deep learning classifier are sometimes seen as mysterious entities. We are delving further into the medical arena of diagnosis, and the ability to explain and comprehension of conclusions are crucial. A more widespread use of these computerized frameworks in the business world is therefore possible. Compared to VGG-16 and Inception, the suggested Ensemble classifier obtains a $100 \%$ accuracy rate, which is much better.