Unstained Blood Smear Analysis: A Review of Rule-Based, Machine Learning, and Deep Learning Techniques.

Abstract

Blood cells are central to oxygen transport, immune defense, and hemostasis. Their number and morphology act as sensitive biomarkers, making accurate segmentation and classification essential for hematological diagnostics. Biophotonic techniques now provide label-free imaging of unstained smears by exploiting intrinsic phase and scattering contrast, yet such images exhibit low optical signal and subtle morphological variation that exacerbate segmentation errors. Label-free modalities nevertheless preserve contrast where dyes fail, motivating renewed interest in unstained workflows. This review analyzes rule-based, machine-learning, and deep-learning approaches for segmenting and classifying label-free blood cells, highlighting performance gains, persistent challenges, and future directions for clinical adoption.