The Utilization of Convolutional Neural Network (CNN) in Leukemia Detection

Abstract

Leukemia is one of the major types of cancer around the world; traditional detection methods of blood smear are time-consuming, subjective, and error-prone. This review provides an overview of the application of Convolutional Neural Networks (CNNs) for detecting leukemia in microscope images. This research investigates the main methods, latest developments, and root causes of the ongoing challenges in CNN-based leukemia detection. This review follows a systematic literature review approach. Grouping recent CNN models in technical routes: (i)pre-training architecture based on transfer learning (e.g., VGG19, AlexNet ) and (ii)dataset augmentation and hyperparameter optimization, against the typical five-step diagnosis workflow. The key findings suggest that CNN significantly improves detection accuracy, speed, and automation. With transfer learning, one of the CNN designs achieved the best results, achieving 99.1% accuracy. Compared with other machine learning methods, accuracy has increased by 2% to 8%. However, most models still suffer from poor dataset quality and diversity, leading to overfitting and poor generaliz ability. This study proposes three complementary solutions: ( i) transfer learning, (ii) unsupervised/semi-supervised learning, and (iii) multimodal data fusion. In conclusion, this review provides a clear roadmap that bridges current technical maps. By integrating three directions, future CNN models can be more robust, low-cost, and clinically viable, thus accelerating the reliable detection of leukemia worldwide.