PNEUMONIA IDENTIFICATION BASED ON THORAX X-RAY USING DENSELY CONNECTED CONVOLUTIONAL NETWORK WITH KERAS AND TENSORFLOW IMPLEMENTATION

• Main Author: Luthfi Aditya, M.
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/60601
• Institution: Institut Teknologi Bandung

The discovery of X-rays has had a huge impact, especially in the medical field in making diagnoses and also therapeutic treatments such as killing cancer cells. In diagnostic use, X-rays can be used to examine patients who have disorders of internal organs. Pneumonia is an inflammation of the lungs caused by bacteria, viruses, or fungi. Pneumonia causes the alveoli in the lungs to become blocked due to mucus fluid in the form of dead tissue or bacteria. In the X-ray procedure, the infected part of the lung is an area of consolidation that will be visible on the X-ray image. By applying the deep learning method, the X-ray image is the amount of data that can be obtained whether it will come on the X-ray image to assist in checking the possibility of having pneumonia or not. By utilizing one of the deep learning methods, namely convolutional neural network (CNN), X-ray images in the form of input can be classified. CNN is a neural network architecture consisting of a matrix that is interconnected between each layer before and before. In this final project, the CNN type used is Densely Connected Convolutional Network with use in image classification in the form of DenseNet-169 architecture. Prediction results for two classes with a learning rate of 10-6 is 88.4% and for three classes prediction is 86.3%. From the program that has been made, it is tested to see the difference in model performance with variations in batch-size. The results obtained show performance with batch-size as 8 giving best results but taking longer when training a neural network model.