Examining a peripheral blood smear under the microscope is usually the first laboratory step when leukemia is suspected, but the examination is slow and depends on the availability of an experienced haematologist. This paper describes a web-based system that supports preliminary screening by classifying microscopic blood smear images into five categories: acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML), chronic lymphocytic leukemia (CLL), chronic myeloid leukemia (CML) and healthy. A small gatekeeper network first rejects uploads that are not blood smear images. Accepted images are classified by a MobileNetV2 network pre-trained on ImageNet and used as a frozen feature extractor, with a trained classification head and a fiveway softmax output. A FastAPI back ed stores every case in MongoDB so that a doctor can review the result before a report is produced. On a held-out test set of 5,000 images (1,000 per class) the classifier reached an overall accuracy of 75.9% and a macroaveraged F1-score of 75.6%, against 97.1% on the training images, which points to overfitting. Healthy smears and CML were recognised most reliably (recall of 99.2% and 94.3%), none of the 4,000 leukemia images was labelled healthy, and most errors occurred among ALL, AML and CLL. The system is meant to assist clinical review and not to replace it.

B G Varsha, Abhinaya, M M Aishwarya, B Sushmita, Vikram sandeep, "AI-Based Leukemia Screening and Blood Smear Analysis System" International Journal of Scientific Research and Engineering Development, V9(5): Page(1769-1773) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.