Fungal Infection Detection Using Custom Fusion-Based Deep Learning Features

Abstract

Fungal infections pose a significant global health challenge, ranging from superficial skin conditions to lifethreatening systemic diseases, particularly in immunocompromised individuals. With an estimated 1.5 million deaths annually, accurate and timely diagnosis is critical. Traditional diagnostic methods, such as microscopy and culture, are often slow and prone to errors. Recent advances in deep learning (DL) have shown promise in analyzing microscopic fungal images for high-accuracy classification. This study focuses on classifying five clinically important fungal species, Candida albicans, Aspergillus niger, Trichophyton rubrum, Trichophyton mentagrophytes, and Epidermophyton floccosum using 5,000 high-quality microscopic images from the publicly available DeFungi dataset. The proposed ensemble approach combining EfficientNet-B1, DenseNet201, and ResNet18 achieved the highest overall accuracy and balanced performance across all classes, particularly improving the classification of Aspergillus niger. These findings highlight the effectiveness of ensemble learning in leveraging diverse feature representations to enhance fungal image classification, offering a reliable and automated diagnostic tool to support clinical decision-making and improve patient outcomes.