Pneumonia Detection Using Convolutional Neural Networks (CNNs)

Pneumonia Detection Using Convolutional Neural Networks (CNNs)

Harsh Shukla, Divyam Tyagi, Rihan Sadiq, Shweta Shekhawat

Under the supervision of: Asst. Prof. Gaurav Singhania
Information Technology,
Noida Institute of Engineering and Technolohy,
Greater Noida, India

Abstract—The main cause of death in children under five is pneumonia, an interstitial lung disease. According to a UNICEF study, it was responsible for about 16% of all deaths among children under the age of five in 2016, killing about 880,000 of them. Most of the affected children were under two years old. Timely identification of pneumonia in children can help to fast-track the process healing. In order to accurately identify pneumonic lungs from chest X-rays, this paper presents convolutional neural network models. These models can be used by medical professionals to treat pneumonia in the real world. The Chest X-Ray Images (Pneumonia) dataset from Kaggle was used for the experimentation.

There are one, two, three, and four convolutional layers in the first, second, third, and fourth models, respectively. The accuracy of the first model is 89.74%, the accuracy of the second is 85.26%, the accuracy of the third model is 92.31%, and the accuracy of the fourth model is 91.67%. In the second, third, and fourth models, dropout regularization is used to reduce overfitting in the fully connected layers. For better evaluation, recall and F1 scores are also computed from each model's confusion matrix.

Keywords— Convolutional neural networks(CNNs), Pneumonia detection, ReLU, Max-pooling, Forward and backward propagation