Digital Morphology for Peripheral Blood Smears in Leukemia Detection: A Systematic Review

Abstract

Background: Peripheral blood smear examination is foundational for leukemia detection, yet manual review is labor-intensive and variable. Digital morphology and artificial intelligence (AI) systems promise faster triage and standardized classification. This review synthesized diagnostic accuracy for detecting leukemia on peripheral smears.

Methods: PubMed was searched from inception to April 2025. Eligible studies were observational diagnostic-accuracy cohorts evaluating digital morphology or AI on peripheral blood smear images against manual microscopy or integrated clinical diagnosis. The primary outcome was sensitivity/specificity; secondary outcomes included predictive values, agreement, and time to result.

Results: Of 1,245 records, 245 duplicates were removed and 1,000 titles/abstracts were screened; 80 full texts were reviewed and 10 cohorts were included. Digital analyzers showed high specificity for common leukocytes (often >90-95%); blast sensitivity varied by platform and case-mix. A compact analyzer reported specificity >94% with blast sensitivity 21-86%. Another platform achieved blast sensitivity 98.4% and specificity 64.0%. AI-assisted APL screening yielded sensitivity 95.8% and specificity 100.0%. Image-classification studies reported sensitivity 97.86% and specificity 100.0% on held-out tests, with APL recall 97.4%. Post-verification correlations for abnormal differentials exceeded 0.93, and PPV/NPV were frequently ≥95%.

Conclusions: Digital morphology and AI reliably triaged peripheral smears with high specificity and context-dependent blast sensitivity. They are best deployed as screening tools with mandatory expert confirmation, supported by local validation and external prospective AI verification.