Computer Aided System (CAS) Of Lymphoblast Classification For Acute Lymphoblastic Leukemia (ALL) Detection Using Various Pre-Trained Models

Computer Aided System (CAS) is an automated, fast and accurate approach for detection and classification purposes. It is used to help experts or medical practitioner as a second opinion to analyze the blood smear image. It is done manually by some practitioners but it is time consuming and creates confusion as different pathologists give different observations and results as it is highly dependent on the experts’ skills. Other than that, it is also challenging to analyze it manually as there are thousands of images. Some researchers used CAS by applying the machine learning to classify the data. However, significant features must be known before proceeding with classification process. In this paper, Convolutional Neural Network (CNN) is applied to classify the WBC types to identify Acute Lymphoblastic Leukemia (ALL). It is a better approach as no complex features need to be designed and it is a fast response program. Pre-trained models of deep learning which are AlexNet, GoogleNet and VGG-16 are compared to each other to find the model that can classify better. There are 260 images in IDB-2 database and 242 images in LISC database. Five types of WBC are classified for LISC database while for IDB-2 database, Lymphoblast and Non-Lymphoblast is classified specifically. As a result, for both database, AlexNet achieve the best result in terms of the training and testing accuracy for each class. Training accuracy for IDB-2 is 96.15% while testing accuracy for Lymphoblast and Non-Lymphoblast is 97.74% and 95.29% respectively. Training accuracy by AlexNet for LISC is 80.82% and testing accuracy is the highest for each class except Monocyte. Overall, AlexNet works better than the other two models for classification for both databases.