Classification of Benign Hematogones and B-cell Acute Lymphoblastic Leukemia in Peripheral Blood Smear Images Using Artificial Intelligence.
Authors
Affiliations (6)
Affiliations (6)
- Internal Medicine, California Northstate University College of Medicine, Elk Grove, USA.
- Osteopathic Medicine, Touro University California, Mare Island, USA.
- Medicine, California Health Sciences University, Clovis, USA.
- Medicine, California University of Science and Medicine, Colton, USA.
- Medicine, University of California, Los Angeles, Los Angeles, USA.
- Internal Medicine, East Tennessee State University Quillen College of Medicine, Johnson City, USA.
Abstract
B-cell acute lymphoblastic leukemia (ALL) is a hematologic malignancy characterized by the abnormal proliferation of immature B-lineage lymphoblasts. Automated analysis of peripheral blood smear images may improve the speed and consistency of cell classification while supporting specialist interpretation. In this study, a cloud-based artificial intelligence image classification model was developed to differentiate benign hematogones from early pre-B, pre-B, and pro-B ALL categories. The model was trained and evaluated using 3,208 microscopic images from a publicly available dataset. Images were divided into training, validation, and testing subsets containing 2,568, 320, and 320 images, respectively, and model development was performed using Google Cloud AutoML (Google LLC, California, US). Performance was evaluated using average precision, precision, recall, precision-recall analysis, confidence-threshold evaluation, and a multiclass confusion matrix. At a confidence threshold of 0.50, the model achieved an average precision of 1.00, a precision of 100%, and recall of 99.4%. Class-specific correct classification rates were 100% for benign hematogones, early pre-B ALL, and pro-B ALL, and 98% for pre-B ALL, with the remaining pre-B images classified as early pre-B ALL. These findings demonstrate strong image-level performance within the internal testing dataset. External, multi-institutional, and patient-level validation is required before the model can be considered for clinical application.