Abstract
Cancer remains a significant global health challenge, with lung cancer being particularly devastating and affecting millions annually. Early detection and accurate classification through computed tomography (CT) imaging are crucial for improving patient outcomes. However, existing approaches face two primary limitations: insufficient training data and suboptimal input modality selection, both of which directly impact model performance.This research addresses these limitations through a novel Dual Modal Classification Approach (DMCA) for lung cancer detection. Unlike traditional single-modal approaches, DMCA integrates information from two distinct sources, CT images and their corresponding segmentation masks, thereby enhancing the model's discriminative power. The underlying premise is that incorporating richer information through multiple modalities optimizes detection performance.
The DMCA was trained and evaluated using the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset. To assess the impact of data availability on model performance and scalability, comparative analyses were conducted using datasets of varying sizes, with equal representation of both image and mask modalities ensured. The fusion of these modalities enabled the DMCA to detect subtle patterns associated with lung cancer more effectively than single-modal approaches.
Our experimental results demonstrate superior performance, with DMCA achieving 91.21% accuracy and 91.18% F1-score on the smaller dataset configuration, and 98.04% accuracy and 98.01% F1-score on the larger dataset. These statistically significant results validate the effectiveness of the dual-modality approach and demonstrate strong scalability characteristics. This research contributes significantly to advancing lung cancer detection methodologies by demonstrating that strategic integration of multiple input modalities can substantially improve both performance and scalability, paving the way for more effective diagnostic tools in clinical settings.
| Date of Award | 19 May 2026 |
|---|---|
| Original language | English |
| Supervisor | Isa Inuwa-Dutse (Main Supervisor) & Qiang Xu (Co-Supervisor) |