Electronic Theses and Dissertation

Citra X-ray paru-paru merupakan salah satu metode diagnostik yang umum digunakan untuk mendeteksi penyakit pernapasan seperti pneumonia dan COVID-19. Namun, kualitas citra sering mengalami degradasi akibat noise, kontras rendah, dan pencahayaan tidak merata, sehingga dapat mempengaruhi performa model deep learning dalam proses klasifikasi. Penelitian ini bertujuan untuk meningkatkan kinerja model MobileNetV2 melalui penerapan metode image enhancement berbasis filtering. Dataset yang digunakan berasal dari RSUD Dr. Zainoel Abidin Banda Aceh dengan total 854 citra yang terdiri dari tiga kelas, yaitu Normal, Pneumonia, dan COVID-19. Tahapan penelitian meliputi enhancement citra, evaluasi kualitas menggunakan PSNR, SSIM, dan entropy, serta klasifikasi menggunakan MobileNetV2. Evaluasi stabilitas dilakukan dengan Performance Statistical Analysis (PSA) melalui lima kali pengulangan eksperimen. Performa model diukur menggunakan accuracy, precision, recall, dan F1-score. Hasil penelitian menunjukkan bahwa kombinasi Highboost Filtering dengan Contrast Stretching menghasilkan akurasi tertinggi sebesar 98%, diikuti Adjust Range dan Contrast Stretching sebesar 96%. Namun, berdasarkan PSA, Contrast Stretching menunjukkan performa paling stabil dengan mean accuracy 96,49% dan standard deviation 0,0037. Hasil evaluasi kualitas citra menunjukkan bahwa PSNR yang tinggi tidak selalu berbanding lurus dengan performa klasifikasi, sedangkan SSIM tinggi dan entropy moderat lebih berkontribusi terhadap peningkatan performa model. Penelitian ini menyimpulkan bahwa peningkatan kualitas citra tidak selalu sejalan dengan akurasi klasifikasi, tetapi bergantung pada kemampuan metode dalam mempertahankan fitur diagnostik. Contrast Stretching memberikan keseimbangan terbaik antara kualitas citra, stabilitas, dan generalisasi model.

Lung X-ray images are a common diagnostic method used to detect respiratory diseases such as pneumonia and COVID-19. However, image quality often degrades due to noise, low contrast, and uneven lighting, which can affect the performance of deep learning models in the classification process. This study aims to improve the performance of the MobileNetV2 model by applying a filtering-based image enhancement method. The dataset used comes from Dr. Zainoel Abidin Regional General Hospital in Banda Aceh with a total of 854 images consisting of three classes: Normal, Pneumonia, and COVID-19. The research stages include image enhancement, quality evaluation using PSNR, SSIM, and entropy, and classification using MobileNetV2. Stability evaluation was carried out using Performance Statistical Analysis (PSA) through five experimental repetitions. Model performance was measured using accuracy, precision, recall, and F1-score. The results showed that the combination of Highboost Filtering with Contrast Stretching produced the highest accuracy of 98%, followed by Adjust Range and Contrast Stretching at 96%. However, based on PSA, Contrast Stretching showed the most stable performance with a mean accuracy of 96.49% and a standard deviation of 0.0037. The results of the image quality evaluation showed that high PSNR was not always directly proportional to classification performance, while high SSIM and moderate entropy contributed more to improving model performance. This study concluded that improving image quality is not always in line with classification accuracy, but depends on the method's ability to retain diagnostic features. Contrast Stretching provides the best balance between image quality, stability, and model generalization.