An effective pyramid vision transformer-based extended depth of focus technique for optimally focused microscopic imaging
JOURNAL OF SUPERCOMPUTING, cilt.82, sa.12, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 82 Sayı: 12
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11227-026-08761-6
- Dergi Adı: JOURNAL OF SUPERCOMPUTING
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Applied Science & Technology Source, Compendex, INSPEC, zbMATH, Academic Search Ultimate (EBSCO), Biomedical Reference Collection: Corporate Edition (EBSCO), Engineering Source (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
Microscopic imaging systems possess an inherent depth of focus (DOF), defining the axial range within which a specimen remains sharply imaged while the detection apparatus remains stationary. When the axial extent of a specimen surpasses this DOF, achieving optimal focus across the entire sample volume becomes infeasible, resulting in only in-focus visualization of regions within the DOF. This limitation can compromise the performance of automated microscopic analysis pipelines. To address this challenge and enhance the efficacy of such applications, extended depth of focus (EDOF) techniques are employed to generate images where the entirety of the specimen appears sharply focused. In this study, we propose a pyramid vision transformer-based EDOF technique (PVT-EDOF) for achieving consistently focused imaging and introduce a novel dataset comprising both real microscopic and synthetically generated image series. The proposed PVT-EDOF technique consists of two primary modules: (1) feature extraction and (2) optimally focused imaging. A key contribution of this work is a focus fusion mechanism that shows promising generalization across the tested microscopes, specimens, and cameras for generating fully focused images. The efficacy of the proposed EDOF technique is quantitatively evaluated using a diverse set of image quality metrics, encompassing both reference-based and no-reference approaches. Comprehensive qualitative and quantitative experimental results demonstrate the superior performance and enhanced feature extraction capabilities of the proposed PVT-based model for EDOF. Since EDOF for high-throughput microscopy must process large focal stacks and gigapixel-scale images, the proposed technique adopts a GPU-accelerated design that processes each image as independent subimages, which makes it inherently parallel and well suited to batched GPU processing and, prospectively, to scaling on high-performance computing resources. Our code and models are publicly available at: .