PRISM: New Segmentation Method for Leukemia Classification
A novel technique known as PRISM (Perinuclear Ring-based Image Segmentation Method) has been introduced for the automated evaluation of Acute Lymphoblastic Leukemia (ALL) in peripheral blood smears. This method circumvents the challenges of traditional membrane-based segmentation, which often suffers from low contrast and variability in cytoplasm. Instead, PRISM creates adaptive concentric zones surrounding the nucleus, enabling the extraction of reliable cytoplasmic descriptors through color and texture statistics, all without the need for precise cell-boundary identification. A calibrated stacking ensemble of conventional classifiers utilizes these descriptors for classification, effectively overcoming the shortcomings of complex neural networks that face difficulties with staining and acquisition variability.
Key facts
- PRISM replaces explicit cytoplasmic delineation with perinuclear concentric zones
- Method extracts descriptors using color information and grey-level co-occurrence texture statistics
- Uses a calibrated stacking ensemble of traditional classifiers
- Addresses low contrast and cytoplasmic variability in blood smear analysis
- Aims to improve generalization across staining and acquisition variability
- Avoids heavy neural architectures and extensive training
- Published on arXiv with ID 2605.12851
- Announce type is cross
Entities
Institutions
- arXiv