2D Lobar Segmentation of the Lung using Convolutional Neural Network

Lung lobe segmentation is essential in medical imaging, as it supports the diagnosis and treatment of respiratory conditions. This research concentrates on segmenting chest CT scans to distinguish five lung lobes: two in the left lung and three in the right lung. We utilized 3D Slicer to manually segment a collection of CT scans, creating ground truth masks for training purposes. To develop a machine learning model, we employed EfficientNet-B0, a cutting-edge convolutional neural network, using PyTorch. The model was trained to perform multiclass segmentation with high precision, tackling the difficulties of segmenting overlapping and low-contrast areas in CT images. To enhance accessibility, we incorporated the trained model into a user-friendly website, enabling users to upload CT scans and receive segmented images in real-time, ensuring accurate predictions. This method highlights the potential for automated lung lobe segmentation in clinical settings. The model was trained on segmented CT scans of the chest region and it showed an accuracy of 83%.