Visual classification of allergenic pollen in iteratively reconstructed lens-less DIHM images

Abstract

Abstract Lens-less digital in-line holographic microscopy (DIHM) is a low-cost, wide-field imaging technique that relies on computational reconstruction to form focused images that should ideally be free of twin-image artifacts. While current DIHM-based pollen classification systems are typically automated and rely on large datasets and deep learning, our study explored whether iteratively reconstructed DIHM images using the Gerchberg–Saxton (GS) algorithm are suitable for visual classification by human experts. Two veterinary cytopathologists evaluated images of six clinically relevant pollen types, namely timothy grass, common ragweed, silver birch, common alder, olive tree, and hazel, using both lens-less DIHM and conventional optical microscopy. Classification accuracy was comparable across modalities, with DIHM achieving 95.8% and optical microscopy 96.9%. Inter-observer agreement was high (Cohen’s κ = 0.91), indicating near-perfect consistency between evaluators. Most misclassifications involved silver birch pollen, likely due to its morphological variability and overlap with common alder and hazel. These findings demonstrate that lens-less DIHM combined with iterative reconstruction enables accurate visual identification of allergenic pollen, offering a promising alternative to conventional microscopy in veterinary and other resource-limited settings.