Deep Learning for analyzing medical images in thoracic diseases is considered necessary in radiology. The thorax examination is the most frequently performed examination in routine examinations. Analytical methods that can explain medical image analysis finding morphological patterns or textures in images are urgently needed. Currently, the interpretation of the thorax is carried out manually by radiologists based on thorax images that are influenced by the cognitive abilities and subjective experience of radiologists, thus allowing human error to occur in reading or analyzing. This study aimed to classify chest diseases such as Edema, Pneumonia, and Pneumothorax using a Convolutional NeurXR) image data used to determine chest disease are normal, edema, pneumonia, and pneumothorax data of 1487, 1022, 1374, and 1128 with an image size of 200×200 pixels. This study also analyzed optimizers RMSprop, Adadelta, Adagrad, Adamax, and Adam in the classification of thorax disease for the CNN algorithm. The results showed that Adam is the best optimizing parameter and can show good accuracy of more than 98% for the classification of thorax diseases such as Edema, Pneumonia, and Pneumothorax with epochs equal to 10.