Automated Mucormycosis Diagnosis from Paranasal CT Using ResNet50 and ConvNeXt Small.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in diagnostics
• Fungal infection detection

Background:

• Mucormycosis is a severe fungal infection requiring prompt diagnosis.
• Traditional diagnosis relies on invasive biopsy, delaying treatment.
• Computed tomography (CT) scans offer a non-invasive imaging modality.

Purpose of the Study:

• To develop and evaluate deep learning models for automated mucormycosis detection.
• To assess the performance of transfer learning models (ResNet50, ConvNeXt Small) on paranasal CT images.
• To determine if AI can expedite mucormycosis diagnosis and aid clinical decision-making.

Main Methods:

• Retrospective analysis of 794 paranasal CT images from patients with mucormycosis, nasal polyps, or normal findings.
• Image preprocessing including resizing and augmentation for model training.
• Fine-tuning of ResNet50 and ConvNeXt Small transfer learning models with a 70/30 train-test split and cross-validation.

Main Results:

• ConvNeXt Small achieved 100% accuracy, precision, recall, and F1-score on the test set.
• ResNet50 demonstrated 99.16% accuracy, with high precision and recall.
• Consistent cross-validation results (~99% accuracy for ConvNeXt) and ablation study confirmed the efficacy of transfer learning.

Conclusions:

• Deep learning models can accurately detect mucormycosis non-invasively from CT scans.
• These AI tools can function as rapid screening aids, complementing histopathology.
• Further validation and clinical integration are recommended for earlier mucormycosis intervention.