Despite good axial resolution in holographic microscopy, lateral resolution remains constrained by optical limitations, pixel size, and noise. These restrictions hinder accurate reconstruction of fine structural and phase details. To address this, we evaluate three deep-learning super-resolution models (RCAN, SwinIR, and a conditional diffusion network) on 1,440 off-axis digital holograms of microbeads downsampled by 2 ×, 3 ×, and 4 ×. We compare their performance to bicubic spline interpolation using PSNR, SSIM, MSE, and phase-derived depth errors. RCAN and SwinIR yield the most accurate reconstructions, preserving structural and quantitative phase information, and offering guidance on model selection for phase-focused holography.

Off-axis holographic microscopy enables single-shot quantitative phase imaging, but its sensitivity is limited by mixed noise that degrades fringe visibility and biases phase recovery. We compare classical filters (Wiener/median/NLM/BM3D and variance-stabilized variants), a supervised Swin transformer, and a diffusion denoiser trained on synthetic data, using 100,000 physics-based holograms with realistic blur, speckle, shot, and read noise. On a 10,000-hologram test set, the best pipeline (diffusion followed by BM3D) achieves 36.86 dB/0.9964 intensity PSNR/SSIM and 46.48 dB/0.9772 phase PSNR/SSIM, reducing RMSE by 88.5% in intensity and 52.8% in phase, with similar improvements on experimental bead and cell holograms.

Dermatophytosis is commonly assessed using potassium hydroxide (KOH) microscopy, yet accurate recognition of fungal hyphae is hindered by preparation-related artifacts, heterogeneous keratin clearance, and notable inter-observer variability. This study presents a transformer-based object detection framework using the RT-DETR architecture for precise, query-driven localisation of fungal structures in high-resolution KOH images. A dataset of 2540 routinely acquired microscopy images was manually annotated using a multi-class strategy that explicitly distinguishes fungal elements from confounding artifacts, enabling the model to actively suppress false detections arising from visually similar mimics. To assess architectural trade-offs, RT-DETR was benchmarked against two CNN-based detectors (YOLOv11 and Faster R-CNN) under identical training and inference conditions. Five-fold stratified cross-validation was performed, and each fold-level model was evaluated on the same independent held-out test set (n = 254). Across the five evaluations, RT-DETR achieved a mean AP@0.50 of $89.73\%\pm 1.48\%$, a mean recall of $0.831\pm 0.011$, and a mean precision of $0.921\pm 0.014$. At the image level, the model achieved a mean sensitivity of $0.989\pm 0.022$ on the independent test set, with a mean of $0.2\pm 0.4$ missed positive cases across the five evaluations. These results demonstrate the technical feasibility of a transformer-based artificial intelligence (AI) system as a decision-support aid for fungal region detection in KOH microscopy, pending prospective multi-center validation to establish clinical generalisability.