A Novel Artificial Intelligence-Driven Technique for Enhancing Medical Imaging Technologies to Aid Diagnosis of Non-Small Cell Lung Cancer

Non-Small Cell Lung Cancer (NSCLC) is a disease wherein malignant cancer cells form in the lung tissue. The five-year survival rate decreases as the NSCLC cancer becomes more advanced, from 40% for stage I to only 1% for stage IV; thus, a vital challenge to overcome is the early and correct detection of NSCLC. Due to radiographers being overworked with many time-consuming duties, new systems that can improve the effectiveness and efficiency of clinical professionals should be considered. Artificial intelligence (AI)-driven techniques can help provide the tools for radiographers to achieve a more accurate and efficient diagnosis of NSCLC. The deep learning (DL) model presented in this study leverages a novel multimodal approach of using both Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) scans as inputs to the model. This allows the model to learn from both morphological data from CT scans and functional, physiological data available from MRI scans. As CT scans are the primary modality used in the structural detection of NSCLC, a higher weight is given to the recommendation made based on the information from the CT scan. One of the most important features analysed by the model is that of the Hounsfield Units (HU) of each pixel within the lung, which are used to pinpoint areas of high density within the lungs that are identified as potential tumours. The model achieved a classification accuracy of 97.1% and an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 95.7% on a test dataset of 140 patients.

URI

https://bradscholars.brad.ac.uk/handle/10454/20731

Type

Thesis

Qualification name

PhD