Electronic Theses and Dissertation

Onychomycosis adalah infeksi jamur pada kuku yang memengaruhi sekitar 10% populasi global dan menjadi salah satu penyakit kuku paling umum dalam praktik klinis. diagnosis konvensional yang memerlukan pemeriksaan laboratorium mendorong kebutuhan akan sistem deteksi yang lebih cepat dan mudah diakses. penelitian ini membandingkan performa dua arsitektur deep learning, yaitu efficientnet-b2 dan vision transformer (vit-base-patch16-224), dalam mengklasifikasikan citra kuku terinfeksi onychomycosis dan kuku normal. dataset yang digunakan berjumlah 18.790 citra dari figshare dan kaggle yang setelah proses pembersihan menghasilkan 16.767 citra bersih, kemudian dibagi secara stratified dengan proporsi 70% data latih, 15% validasi, dan 15% data uji. kedua model dilatih menggunakan pendekatan two-phase fine-tuning dengan transfer learning dari pretrained weights imagenet. hasil evaluasi pada 2.516 gambar uji menunjukkan bahwa efficientnet-b2 mencapai akurasi 97,93%, presisi 97,95%, recall 97,91%, dan f1-score 0,9793, sedangkan vit-base-patch16-224 mencapai akurasi 97,69%, presisi 97,69%, recall 97,69%, dan f1-score 0,9769. hasil uji statistik paired t-test menunjukkan bahwa perbedaan performa kedua model tidak signifikan secara statistik (p > 0,05), namun efficientnet-b2 tetap menjadi pilihan yang lebih optimal karena memiliki jumlah parameter yang jauh lebih sedikit (~9 juta) dibandingkan vit-base (~86 juta parameter). model terbaik diimplementasikan dalam prototipe aplikasi web berbasis flask yang dapat digunakan untuk mendeteksi onychomycosis secara otomatis berdasarkan citra kuku yang diunggah oleh pengguna. kata kunci : onychomycosis, efficientnet-b2, vision transformer, deep learning, transfer learning, klasifikasi citra onychomycosis.

Onychomycosis is a fungal nail infection that affects approximately 10% of the global population and is one of the most common nail diseases in clinical practice. Conventional diagnosis requiring laboratory examinations drives the need for faster and more accessible detection systems. This study compares the performance of two deep learning architectures, namely EfficientNet-B2 and Vision Transformer (ViT-Base-Patch16-224), in classifying images of onychomycosis-infected nails and normal nails. The dataset used consisted of 18,790 images from Figshare and Kaggle. After the cleaning process, 16,767 clean images were obtained. These images were then divided stratified with a proportion of 70% training data, 15% validation data, and 15% test data. Both models were trained using a two-phase fine-tuning approach with transfer learning from pretrained ImageNet weights. The evaluation results on 2,516 test images showed that EfficientNet-B2 achieved 97.93% accuracy, 97.95% precision, 97.91% recall, and F1-score of 0.9793, while ViT-Base-Patch16-224 achieved 97.69% accuracy, 97.69% precision, 97.69% recall, and F1-score of 0.9769. The paired t-test results showed that the difference in performance between the two models was not statistically significant (p > 0.05), but EfficientNet-B2 remains a more optimal choice because it has a much smaller number of parameters (~9 million) than ViT-Base (~86 million parameters). The best model was implemented in a Flask-based web application prototype that can be used to automatically detect onychomycosis based on user-uploaded nail images. Keywords : Onychomycosis, EfficientNet-B2, Vision Transformer, Deep Learning, Transfer Learning, Onychomycosis Image Classification.