Detection of Acute Lymphoblastic Leukemia Using a Novel Bone Marrow Image Segmentation

This project presents a comprehensive image processing and classification approach for the detection of Acute Lymphoblastic Leukemia (ALL) using segmented bone marrow images. The methodology begins with the acquisition of bone marrow images, followed by preprocessing steps including grayscale conversion, contrast enhancement, and noise removal through median filtering. Thresholding techniques are applied to segment potential leukemic regions, and morphological operations such as closing, dilation, and erosion further refine the segmentation. Small objects are eliminated to reduce noise and improve clarity of the segmented cell regions. A two-dimensional Haar wavelet transform is applied to extract approximation coefficients from the segmented image, which are then used for Gray Level Co-occurrence Matrix (GLCM)-based texture feature extraction. Key features such as contrast, homogeneity, mean intensity, energy, area, and eccentricity are computed to characterize the cellular structures. These features form the input vector for classification. An Adaptive Neuro-Fuzzy Inference System (ANFIS) model is trained using a dataset labeled with six different cell types, including Band cell, Metamyelocyte, Myeloblast, N. Myelocyte, N. Promyelocyte, and Neutrophil cell. The trained ANFIS model predicts the cell type from the extracted features of the input image. Finally, the classification performance is evaluated using metrics such as accuracy, precision, sensitivity, specificity, and F1 score, demonstrating the effectiveness of the proposed method. This automated system offers a promising computer-aided diagnostic tool to support hematologists in the early detection and classification of ALL, thereby improving diagnostic accuracy and reducing the workload of medical professionals.

Keywords: Dataset, Deep Learning, Convolution Neural Network, Image Processing Techniques, segmentation and accuracy.