Evaluation of key focus measures for smear analysis

Abstract

Reliable focus assessment is crucial in microscopic smear analysis, as automated systems' diagnostic accuracy is heavily reliant on image sharpness. Despite the availability of numerous focus measures (FMs), their applicability across biological domains remains unknown. This study employs a weighted, raw value-based generalization approach to evaluate fifty single focus measures across five publicly available Z-stack microscopy datasets. Ten quantitative criteria were used to assess focus curve fidelity and resilience. The raw-value-based generalization score combines weighted mean performance and inter-dataset stability to provide a complete measure of cross-domain reliability. Many FMs have been proposed, including gradient, statistical-, texture-, and transform-based operators. However, their efficacy varies greatly depending on specimen type, staining process, and optical arrangement. Prior comparative studies [4], [5] typically focused on one or two datasets or imaging modalities, limiting the extent to which their recommendations may be applied to larger smearanalysis contexts. Recent research has also offered deep learning-based focusing algorithms for digital pathology and whole-slide imaging [5]. While effective, these systems typically require task-specific training data, significant computational resources, and function as black box predictors rather than interpretable attention scores. In contrast, traditional handcrafted FMs remain appealing for incorporation with resource-constrained or older autofocus systems that require transparency, minimal latency, and ease of deployment. Our study presents a comprehensive evaluation of fifty single-focus measures from five publicly available z-stack microscopy datasets that include haematological and microbiological smears. A raw-value-based generalization methodology is suggested to measure both performance and inter-dataset stability, with the goal of discovering a limited set of FMs that are consistently dependable across a wide range of smear types and imaging conditions. The collected findings assist researchers and developers in identifying the best focus operators for future biomedical imaging applications.