<p>Leukemia is a life-threatening hematological malignancy that affects white blood cells and bone marrow. It demands efficient, timely and accurate diagnostic procedures to improve patient outcomes. Acute Lymphoblastic Leukemia (ALL) is a common type of cancer in pediatric populations. This research undertakes a systematic review of deep learning (DL) techniques for identifying and categorizing ALL through the use of blood smear imaging modalities. In accordance with PRISMA guidelines, a bibliometric analysis of DL-based studies published from 2019 to 2025 is undertaken. This analysis investigates methodologies, datasets, and performance trends in the detection of ALL. The review underscores the importance of preprocessing strategies, feature extraction pipelines, and classification frameworks of different methods. The methods are convolutional neural networks, transfer learning models, attention mechanisms, combined CNN–Transformer structures, and Explainable AI techniques. Widely used benchmark datasets such as ALL-IDB, C-NMC, and BloodMNIST are examined with respect to their usage patterns and reported performance. Many studies show high accuracy in controlled experiments. This review highlights ongoing challenges, such as limited dataset size and diversity, class imbalance, overlapping cellular morphology, and inadequate external clinical validation. The findings suggest that while DL has demonstrated significant potential for automated ALL diagnosis, existing systems are in the research and prototype stages. Considerable work is required to create reliable and affordable medical solutions. The review identifies important research gaps and suggests future steps. It highlights the necessity for diverse and clinically representative datasets, enhanced generalisation, and the integration of explainable models. The focus is also on creating lightweight architectures that are suitable for real-world clinical deployment.</p>