Abstract:
Acute Lymphoblastic or lymphocytic leukemia (ALL) is a lethal type of haematological malignancy of white blood cells in which abnormal white blood cells are
produced.These abnormal cells affect the whole blood circulatory and immune
systems and spongy tissues of the bone-marrow. The prevalence rate of the acute
leukemia is very high. Usually the malignant cells proliferate and replace most
of the normal healthy cells in body. ALL can be very fatal if not treated and
controlled in early stages.Diagnosis of the disease involved in-depth symptoms
analysis followed by methods and procedures of pathological investigations.These
investigations are carried out by experienced professionals according to the WHO
and FAB standards.Diagnosis and prognosis of ALL is still a major issue in devel oping countries due to lack of trained medical professionals and costly diagnostic
instrumentation. Therefore, A low cost, easy to use automated computer based
tool for the detection and classification of the disease has the capability to decrease
the mortality rate of ALL. However, classification of such cells in microscopic image is very challenging due to the morphological similarities of the abnormal and
healthy cells. This work is based on the efforts made for the development of an
automatic classification tool for ALL. We investigate different techniques for the
task and exploited deep learning based methods. Specifically we use combination
of CNNs and RNNs. Moreover, we exploited frequency domain dynamic features,
texture features and stain de-convolution based optical domain transformation for
improved performance. An ensemble model of different existing and custom models (utilizing the mentioned features) is created for the classification task. Our
proposed architecture outperformed the existing state of the art deep learning architectures and achieved reasonable classification accuracy.
