A NOVEL machine learning approach to classifying acute lymphoblastic leukaemia (ALL) has been developed by researchers. Timely and accurate detection of ALL is critical for effective treatment and management. Recent advancements in computer-assisted diagnosis (CAD) systems aim to support hematologists by streamlining workload and handling large datasets. Despite advances in computer-aided leukaemia classification, applying AI to clinical diagnosis is challenging due to complex variations in blood cell images. Traditional machine learning and deep learning models often disregard extraneous information and face overfitting issues, leading to poor generalisation. Additionally, certain models struggle with redundant computation and gradient disappearance, and excessive pooling operations result in the loss of global and local context information. These issues can lead to inaccurate classification.
Therefore, researchers used a Deep Dilated Residual Convolutional Neural Network (DDRNet) to enhance the classification of blood cell images, specifically focusing on eosinophils, lymphocytes, monocytes, and neutrophils. This model addresses several challenges mentioned above, such as vanishing gradients and the need for effective feature extraction.
The DDRNet model was tested using a Kaggle dataset of 16,249 images. Among these, 12,515 images were utilised for training and validation, consisting of 3,133 eosinophil images, 3,109 lymphocyte images, 3,102 monocyte images, and 3,171 neutrophil images. The remaining 3,734 images were reserved for testing, which comprised 936 eosinophil images, 931 lymphocyte images, 927 monocyte images, and 940 neutrophil images.
The experimental results demonstrated that the DDRNet model achieved an impressive classification accuracy of 99.86% for training data and 91.98% for testing data. The model had an F1 score of 0.96, demonstrating its feature discrimination capability with minimal computational complexity.
In summary, the DDRNet model excels in classifying blood cell images related to ALL, by addressing key challenges in the classification process, significantly enhancing the model’s accuracy and feature discrimination abilities. The DDRNet model represents a significant advancement in CAD systems for leukaemia diagnosis, offering superior performance compared to existing methods. Future research should focus on the development of AI models for enhanced classification of other types of leukaemia.
Katrina Thornber, EMJ
Reference
Jawahar M et al An attention-based deep learning for acute lymphoblastic leukemia classification. Sci Rep. 2024;14(1):17447.
