Scholarlink

<p>Acute Lymphoblastic Leukaemia (ALL) is a serious hematologic malignancy. It is imperative to diagnose the disease quickly. Automatic detection may facilitate the process. In the proposed work, the YOLOv8 object detection framework is used to detect the four distinct types of B-cells of ALL accurately by analyzing the images of Peripheral Blood Smear (PBS) using the deep learning technique. The technique is discussed in the proposed work. The existing method includes experimental validation, object detection/classification, performance evaluation, and data preprocessing/augmentation. The first step in the preparation pipeline is to ensure that all photos are 640 by 640 pixels in size. Then, normalisation and noise reduction using the Gaussian Blur filter are used. In order to enhance the model's capacity to generalise to new images with varying orientations, certain preprocessing processes, such as arbitrary horizontal and vertical twisting and rotation, are implemented. The significance of this work is the examination of the proposed model's ability to perform multi-level classification, not only between Normal (Benign) and Abnormal (Malignant) cells, but also to further classify malignant cells into specific subtypes, including Early Pre-B cell, Pre-B cell, and Pro-B cell structures. Additionally, it transmits a confidence score for each cell identified to enhance the reliability of the detection and provides the absolute cell count in each PBS image, thereby aiding medical professionals and physicians in assessing the severity of the disease. The model training process yielded exceptional performance metrics, such as precision of 96%, recall of 95%, Mean Average Precision (mAP) @0.50 of 98%, mAP@[0.5:0.95] of 74%, and accuracy of 99.3%. These results demonstrate the significant potential of YOLOv8 to serve as a fast and dependable tool for in-depth ALL subtyping on multi-cell blood smear images, thereby making a significant contribution to the field of automated haematological diagnostics. A significant confidence score of predicted class has been demonstrated in experimental validation on unobserved sample images.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

YOLOv8-based detection, subtyping, and cell counting of B-cell acute lymphoblastic leukemia in peripheral blood smear images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Soni Mayank,
  • Suryansh Ambekar,
  • Usha Verma,
  • P. Chandra Babu,
  • S. Narasimha Raju Akella,
  • Ranith Kumar Gatla,
  • A. Almenweer Reem

摘要

Acute Lymphoblastic Leukaemia (ALL) is a serious hematologic malignancy. It is imperative to diagnose the disease quickly. Automatic detection may facilitate the process. In the proposed work, the YOLOv8 object detection framework is used to detect the four distinct types of B-cells of ALL accurately by analyzing the images of Peripheral Blood Smear (PBS) using the deep learning technique. The technique is discussed in the proposed work. The existing method includes experimental validation, object detection/classification, performance evaluation, and data preprocessing/augmentation. The first step in the preparation pipeline is to ensure that all photos are 640 by 640 pixels in size. Then, normalisation and noise reduction using the Gaussian Blur filter are used. In order to enhance the model's capacity to generalise to new images with varying orientations, certain preprocessing processes, such as arbitrary horizontal and vertical twisting and rotation, are implemented. The significance of this work is the examination of the proposed model's ability to perform multi-level classification, not only between Normal (Benign) and Abnormal (Malignant) cells, but also to further classify malignant cells into specific subtypes, including Early Pre-B cell, Pre-B cell, and Pro-B cell structures. Additionally, it transmits a confidence score for each cell identified to enhance the reliability of the detection and provides the absolute cell count in each PBS image, thereby aiding medical professionals and physicians in assessing the severity of the disease. The model training process yielded exceptional performance metrics, such as precision of 96%, recall of 95%, Mean Average Precision (mAP) @0.50 of 98%, mAP@[0.5:0.95] of 74%, and accuracy of 99.3%. These results demonstrate the significant potential of YOLOv8 to serve as a fast and dependable tool for in-depth ALL subtyping on multi-cell blood smear images, thereby making a significant contribution to the field of automated haematological diagnostics. A significant confidence score of predicted class has been demonstrated in experimental validation on unobserved sample images.