Lung Cancer Detection Using Positron Emission Tomography Images Through Convolutional and Recurrent Neural Networks

Lung cancer ranks third in cancer detection, following breast and prostate cancer, but it causes more deaths compared to other cancer types. It is divided into Non-Small Cell Lung Cancer (NSCLC), comprising three sub-types, and Small Cell Lung Cancer (SCLC). Distinguishing between these two types is challenging for doctors, necessitating the development of new technological tools. This work introduces a two-phase method with a total of eight steps to identify the sub-types of NSCLS (adenocarcinoma and squamous cell carcinoma) and SCLC. The process involves gathering and loading data, selecting PET scans, transforming them into RGB images, creating patient videos, and applying data augmentation before splitting the data into training and testing sets. Three models are built using Convolutional Neural Networks with hyperparameter tuning, VGG16, and ResNet50, each assembled with a Gated Recurrent Unit (GRU). The models are trained and tested, and the results are presented. The method utilizes a public dataset from The Cancer Imaging Archive, specifically the Lung-PET-CT-Dx dataset. Moreover, this approach can potentially be extended to detect and predict other types of cancers and diseases based on PET scans.