Automated microbial recognition from gram-stained micrographs via CNN and metadata decision support
Frontiers in Cellular and Infection Microbiology, cilt.16, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16
- Basım Tarihi: 2026
- Doi Numarası: 10.3389/fcimb.2026.1811365
- Dergi Adı: Frontiers in Cellular and Infection Microbiology
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, EMBASE, MEDLINE, Directory of Open Access Journals, Zoological Record, Natural Science Collection (ProQuest), Biological Science Database (ProQuest), Health Research Premium Collection (ProQuest)
- Anahtar Kelimeler: bacterial identification, convolutional neural network, deep learning in microbiology, gram stain, metadata integration, microscopy image classification
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
Introduction – Accurate and timely identification of microorganisms is critical for guiding clinical decision-making, ensuring biosafety, and supporting microbiological research. Standard morphological assessments often require significant manual effort and specialized expertise, underscoring the need for automated, high-throughput diagnostic tools. This study presents an automated species-level bacterial classification pipeline that couples deep learning analysis of Gram-stained microscopy images with a clinical decision-support framework fueled by a curated microbiological metadata repository. Methods – A convolutional neural network (CNN) was trained on 2, 034 brightfield microscopy images representing 33 bacterial species of clinical and industrial relevance across diverse taxa and imaging conditions. Preprocessing protocols included intensity normalization, spatial resizing, and on-the-fly data augmentation to ensure generalization. The preprocessed inputs were evaluated using a sequential CNN architecture optimized for multiscale feature extraction. The pipeline subsequently linked visual model predictions to an integrated repository containing eight laboratory-relevant metadata attributes per species, including Gram status, cellular morphology, oxygen requirement, biosafety level, and pathogenicity profiles. Results – On an independent test set, the classification model demonstrated balanced diagnostic capability, achieving an accuracy of 0.84, a weighted F1-score of 0.84, and a Matthews correlation coefficient of 0.84 across all target classes. Morphologically distinctive species, including Actinomyces israelii, Candida albicans, and Neisseria gonorrhoeae, were classified with perfect precision and recall (1.00). Conversely, species exhibiting high intra-genus morphological overlap, particularly within the Lactobacillus genus, demonstrated comparatively lower classification metrics due to shared phenotypic traits under light microscopy. Discussion – Coupling deep learning classification with contextual microbiological metadata yields standardized, interpretable diagnostic reports that bridge the gap between pure image recognition and actionable clinical intelligence. While morphological convergence remains a limitation for certain closely related taxa, the integration of structured metadata offers a reliable decision-support mechanism for laboratory and educational environments. This automated framework establishes a foundation for multimodal diagnostic pipelines that can incorporate biochemical assays, spectroscopic data, and expanded clinical cohorts in prospective validation studies.