Inception-ResNet-v2 with Leakyrelu and Averagepooling for More Reliable and Accurate Classification of Chest X-ray Images

Pneumonia is one of the most commonly seen illnesses in the world and its diagnosis needs some expertise. Computer aided diagnosis methods are used extensively in a lot of fields like health care. This study uses Inception-ResNet-v2 deep learning architecture. Classification is done by using this architecture. ReLU activation function seen in network architecture is changed with LeakyReLU activation function and classification task is done. After that, all of the maxpooling layers seen in network architecture is changed with avepooling layers and again classification task is done. Lastly, this seperate changes done in network architecture is combined in one network and again classification task is done with new network architecture. Four experiments are done in total and their results are compared. The best case with a sensitivity value of 93.16% and with a specificity value of 93.59% is obtained in Inception Resnet V2 with together application of LeakyReLU and Averagepooling.