Multi-layer Convolutional Approach for Lung Cancer Detection using CXR

Lung cancer is considered as one of the most significant causes of deaths globally. Diagnosis at an initial stage, using computed tomography chest scans could give a better chance to the patient to survive by providing an opportunity for effective care plans and treatment. We propose a new deep-learning method to learn high level image representation towards attaining a significant classification accuracy. This technique consists of three important steps, which are data preparation, pre-processing of data, and classification. The proposed model is a multi-layer convolutional neural network (CNN) that uses different convolutional layers, pooling layers, flatten, dense layers, dropout layers, and performs classification. Two pretrained models which are VGG16 and Densenet, that takes weights using ImageNet pretraining are also employed. This work utilized chest CT-scan image dataset. The dataset contains the images in PNG or JPG format which are suitable for the model. The data contains three types of chest cancers which are adenocarcinoma, large cell carcinoma, and squamous cell carcinoma as well as normal controls. Our experimental results showed that the proposed models achieved the maximum accuracy of 99.75% using the multi-layer CNN model, of 97.25% using Densenet-201, and of 96% using VGG-16.