Indah Purnama Sari, Al-Khowarizmi, Fanny Ramadhani, Andy Satria
Tuberculosis (TB) is an infectious disease caused by the bacterium Mycobacterium tuberculosis, a pathogen that most successfully infects the lungs. The most frequently used technique for diagnosing TB disease is through x-ray examination of the thorax (lungs). The X-ray results can be seen image/visually by the doctor to see whether there are characteristics and patterns of TB disease in the patient. This research focuses on Artificial Intelligence (AI) to help doctors and provide efficient alternative solutions in diagnosing patients, whether the patient is suffering from TB or not more quickly. This research was designed using a Multi-Scale Convolutional Neural Network (CNN) to classify tuberculosis based on chest x-ray images. The data used is an X-ray image of the thorax which is used as input for the image processing process. The dataset collected amounted to 1400 data consisting of 2 classes, namely normal lungs and lungs of tuberculosis sufferers. The CNN model consists of 3 convolution layers measuring 3x3, 3 pooling layers (Maxpool) measuring 2x2 and 1 fully-connected layer that uses softmax activation. The filters used in each convolution layer are 128. This research uses the Adam Optimizer algorithm. The dataset is divided into 1120 data in scenario 1 and 978 data in scenario 2 for training and 280 data in scenario 1 and 422 data in scenario 2 for testing. In the training process, the epoch value of 20 was used to obtain an accuracy value of 100% in all scenarios. In the testing stage, an accuracy value of 99.29% was obtained in scenario 1 and 97.67% in scenario 2. © 2024 IEEE.
Universitas Muhammadiyah Sumatera Utara, Department of Information Technology, Medan, Indonesia; Universitas Negeri Medan, Department of Computer Science, Medan, Indonesia; Universitas Dharmawangsa, Department of Information Technology, Medan, Indonesia
