Metaheuristic Optimization Based Deep Learning Models for Classification of Leukemia from Blood Microscopic Images

Abstract

Detecting the hematological leukocyte structures is considered a complicated task due to the enhanced mortality rate and high number of new cases. In recent years, hematological specialists utilize microscopic investigations of the peripheral blood smear for diagnosing various hematological deficiencies. Leukemia is typically considered to be caused by a sudden increase in the total number of abnormal or immature cells. The majority of the new blood cells produced daily by the bone marrow are red cells. But leukemia - infected body generates more white cells than required. AML commonly affects immature cells and the symptoms are fever, lethargy, hemorrhage, fatigue and anemia. Some risk factors are exposure to radiation and chemicals, having a genetic disorder, blood disorder, being subjected to other cancer treatments like radiation therapy, etc. newlineIntermittently, the AML cells generate solid tumor cells referred to as chloroma or myeloid sarcoma that can be developed everywhere in the human body. During recent years, numerous technical advances employing cytogenic detection and molecular approaches have generated a novel insight into both the biological and genetic features of acute myeloid leukemia. On the other hand, due to inaccurate results, complexity, high cost and more time the techniques become less newlinevi newlineeffective. The underlying motivation of this thesis is to obtain an optimal routing path using three significant phases. In the initial phase, we utilized two approaches, namely the fractional black widow optimization algorithm and neural network to detect AML. First, the noise and various redundant signals are eliminated from the input phase in the pre-processing phase. The Adaptive Fuzzy Entropy (AFE) strategy, which combines the adaptive counter-based technique and the fuzzy C-means clustering approach, is then used to segment the AML defective region.