ENSEMBLE MODEL WITH IMPROVED U-NET-BASED SEGMENTATION FOR LEUKEMIA DETECTION

Abstract

An essential component of the immune system that aids in the fight against pathogens is white blood cells. One of the most prevalent blood diseases, leukemia can be fatal if not properly diagnosed. Diagnosing this disease at an early stage may reduce the severity of the disease. This research intends to propose an ensemble model with improved U-net for leukemia detection (EMIULD) with the following four phases: preprocessing, segmentation, feature extraction and detection. The preprocessing step involves preprocessing the blood smear image, which includes filtering and scaling the image. The segmentation phase is applied to the preprocessed image, and U-Net-based segmentation is used to segment the image. As a result, features for the segmented images are extracted, including better Local Gabor XOR Pattern (LGXP), area, and grid-based form features. The extracted features are fed into the suggested ensemble model, which consists of Deep Convolutional Neural Network (DCNN), Support Vector Machine (SVM) and Random Forest (RF) classifiers, with the purpose of detecting leukemia. Finally, the proposed Bidirectional Long Short-Term Memory (Bi-LSTM) network to predict whether the given blood smear image is leukemia or not. The suggested model attained the best outcome when evaluated over the extant approaches.