Developing a CAD System to Detect Pulmonary Nodules from CT-Scan Images via Employing 3D-CNN

Due to high development in machine learning with satisfactory results in medical image detection and segmentation approaches, Several Computer-Aided Diagnosis (CAD) systems are adopted to help detect and diagnose pulmonary lung nodules. This paper proposes two CAD systems (Model-1 and Model-2) to classify benign or malignant tissue by adopting a three-dimension Convolution Neural network (3D-CNN) with a 3D-CT scan image. Initially, a seven convolutional layer was adopted in the first model (Model -1), with one fully connected layer. In terms of accuracy, the first proposed model outperformed the current state-of-the-art by a significant margin (98.9 percent). A block of convolution and max- pooling layers known as the inception layer is employed in the second model (Model -2). Model -2 is developed with two convolution layers and four inceptions layers to train a dense convolution neural network followed by one fully connected layer to detect malignant or benign tissue accurately. The second proposed model achieved state-of-the-art performance and significantly outperformed in accuracy levels of around (99.5%). Finally, the proposed Model (Model -2) performance is compared with some related work that has applied the same dataset or utilized a different dataset and gives a higher performance with classification accuracy reach to 99.5 %. It's also worth noting that sensitivity and specificity came out on top compared to other studies, with a 99.8 and a 99.1 percentage, respectively.