Prediction of Pneumonia from 3D CT Images Employing Deep Neural Network based Feature Extraction Method

Pneumonia is an inflammatory lung infection that can cause life-threatening consequences such as difficulty breathing, fever, and chest pain. Early detection enables timely action, which reduces the severity of the ailment and improves patient outcomes. However, because of the complexity and diversity of its indications, pneumonia prediction poses considerable problems, making accurate diagnosis and prognosis difficult. In this work, a novel 3D feature extraction method for pneumonia prediction from computed tomography (CT) image volume is proposed. A 3D convolution neural network with rigorous pre-processing processes is used to construct the feature extraction method. The obtained features are then utilized for making decisions in which a number of layers are put together to develop the pneumonia prediction model. A 3D CT image dataset with 160 volumes and 40 volumes for training and testing respectively were used for the experimental evaluation. Several performance indicators were employed to evaluate performance, and cutting-edge methods were utilized to compare results. From the experiment, the proposed method achieved 95% training and 85% validation accuracy in which the state of art 3D CNN models accuracy reached up to 88-92% and 76-80% respectively. In conclusion, it was shown that the suggested algorithm outperforms others in terms of predicting pneumonia from 3D CT volume.