DNN-ILD: A Transfer Learning-Based Deep Neural Network for Automated Classification of Interstitial Lung Disease from CT Images

Interstitial lung disease (ILD) is a term commonly used to refer to several types of lung disorders that are harmful to humans. ILD is challenging to categorize. Analyzing the type of ILD from computed tomography (CT) images by radiologists is very time-consuming. Computer-aided diagnostic (CAD) technology has been constructed to excite detection. This work introduces one such method that relies on convolutional neural networks (CNN) to detect ILD from chest CT images. We examined domain-specific neural networks and pre-trained neural networks for problem-solving. We introduced various approaches to enhance a CNN’s capability in categorizing ILD. We have proposed a new deep neural network (DNN) based on the modified DenseNet121 with hyper-parameter tuning and fivefold cross-validation. The performance of the proposed model is measured using accuracy and confusion matrix. The test accuracy of four types of detection (pulmonary fibrosis, hypersensitivity pneumonitis, tuberculosis, and healthy lung) on ResNet50, VGG16, and modified DenseNet121 models are 62%, 86.2%, and 89.6%, respectively. The proposed method has achieved better training, validation, and test accuracy on the MedGIFT ILD dataset and outperforms the state-of-the-art models.