<![CDATA[Diabetes mellitus is a chronic metabolic disease that requires early detection to prevent complications. However, commonly used diagnostic methods are still invasive and require laboratory testing. This study aims to develop a diabetes detection model based on nail image analysis using the Hybrid Vision Transformer–Convolutional Neural Network–Long Short-Term Memory (Vision Transformer–CNN–LSTM) method as a non-invasive approach. The study included dataset collection, image preprocessing, including resizing, normalization, segmentation, dataset partitioning, model training, and evaluation using a confusion matrix. The Vision Transformer was used to capture a global image representation, the Convolutional Neural Network extracted local features, and the Long Short-Term Memory enhanced the feature representation before the classification process. Test results showed that the model achieved 93.33% accuracy, 91.89% precision, 94.44% recall, 93.13% F1-score, and an Area Under the Curve of 0.972. These results demonstrate that the proposed model is capable of accurately detecting diabetes and has the potential to be a fast, easy, and non-invasive alternative for initial screening based on nail images.Keywords: Diabetes mellitus; nail image; Vision Transformer; Convolutional Neural Network–Long Short-Term Memory; Early detection. AbstrakDiabetes mellitus menjadi penyakit metabolik kronis yang memerlukan deteksi dini untuk mencegah terjadinya komplikasi, namun metode diagnosis yang umum digunakan masih bersifat invasif dan memerlukan pemeriksaan laboratorium. Penelitian ini bertujuan mengembangkan model deteksi diabetes berbasis analisis citra kuku menggunakan metode Hybrid Vision Transformer–Convolutional Neural Network–Long Short-Term Memory (Vision Transformer–CNN–LSTM) sebagai pendekatan noninvasif. Penelitian dilakukan melalui tahapan pengumpulan dataset, preprocessing citra berupa resize, normalisasi, segmentasi, pembagian dataset, pelatihan model, dan evaluasi menggunakan confusion matrix. Vision Transformer dimanfaatkan untuk menangkap representasi global citra, Convolutional Neural Network mengekstraksi fitur lokal, sedangkan Long Short-Term Memory memperkuat representasi fitur sebelum proses klasifikasi. Hasil pengujian menunjukkan bahwa model menghasilkan accuracy 93,33%, precision 91,89%, recall 94,44%, F1-score 93,13%, dan Area Under Curve sebesar 0,972. Hasil tersebut menunjukkan bahwa model yang diusulkan mampu mendeteksi diabetes secara akurat serta berpotensi menjadi alternatif skrining awal berbasis citra kuku yang cepat, mudah, dan noninvasif.Kata kunci: Diabetes mellitus; citra kuku; Vision Transformer; Convolutional Neural Network–Long Short-Term Memory; Deteksi dini]]>
Copyrights © 2026
