Abstract:
Multiple myeloma is a haematological cancer that occurs in plasma cells. The increase in the myeloma cells induces the reduction in all the blood cells. Production of abnormal antibodies or multiple myeloma proteins (for example, IgG or IgA) is the consequence of multiple myeloma cells. Myeloma cells in bone marrow images are detected automatically in the current study. Various steps including preprocessing, feature extraction and segmentation are used to detect myeloma cells in bone marrow because microscopic examination is not sufficient enough. This inadequacy is because of involvement of human factors i.e., fatigue, stress, proficiency, experience etc. Therefore, there is utmost need of some automated method to detect myeloma cells in bone marrow aspirations to achieve accuracy in performance, and making the entire process timeeffective. Convolutional neural network is employed in order to detect multiple myeloma while using microscopic images of blood. Pre- trained and fine- tuned AlexNet is taken into consideration instead of developing an architecture from scratch. Input images are classified into normal and blast by substituting last layer of pre- trained AlexNet by two new layers. Besides, there is the need of classifier in order to classify myeloma cells. Hence, SVM is used for this purpose. Purposed methodology was compared with the previous literature in terms of total number of images in the data set and their performance as well. Our methodology unlike previous ones performed well in terms of carrying out 100 % accuracy without employing any microscopic segmentation.
