SENSEI: Classification of Lung Disease with Convolutional Neural Network Method on X-ray Image

Lung disease is one of the most common and serious health disorders worldwide. The condition causes the lungs to be unable to function normally, with most cases caused by exposure to cigarette smoke. Diagnosis of lung disease currently relies heavily on X-ray examinations by lung specialists. However, this method does not always provide accurate results, so it is recommended to perform a series of additional tests to improve the accuracy of the diagnosis. This research focuses on the classification of lung X-ray images using Deep Learning techniques to detect various types of lung diseases, including Pneumonia, Tuberculosis, and Corona virus Disease-19 (Covid-19). The method used is Convolutional Neural Network (CNN), which is the most significant Deep Learning method currently in medical image recognition. Furthermore, a combination of Hyperparameter (epoch, mini batch size, dataset variation) and Optimizer (Adam, SGDM, RMSProp) is tested to find the optimal model with the highest accuracy and lowest loss value. This study uses a dataset of 1,050 images divided into 3 classes, namely Pneumonia, Tuberculosis, and Covid-19. Each class consists of 350 image data. The test results show that the CNN model can achieve up to $100 \%$ accuracy in the classification of lung diseases in X-ray images. This best accuracy is obtained by using the RMSPROP and Adam optimizers. Tests were conducted with a variety of datasets $(\mathbf{6 0 \%}, 70 \%, 80 \%$, and $90 \%)$, mini batch sizes $(8,16,32$, $64)$, and the number of epochs $(10,20,25,30)$. The final results show that the optimal combination of data, Optimizer, and model parameters is $\mathbf{6 0 \%}, \mathbf{7 0 \%}, \mathbf{8 0 \%}, \mathbf{9 0 \%}$ data division with Adam & RMSPROP Optimizer, epoch 15, 20, 25, 30 and mini batch size 8.