Lung cancer is the leading cause of cancer-related deaths globally, with early and accurate detection being crucial for patient survival. While computed tomography (CT) scans are instrumental in identifying pulmonary nodules, current deep learning models typically address segmentation and malignancy classification as separate tasks, potentially overlooking the mutual benefits of joint optimization. In this work, we propose MTL-LungNet, a novel multi-task learning framework that simultaneously performs nodule segmentation and malignancy classification using CT data. By leveraging shared representations and a segmentation-guided classification strategy, our model captures both spatial precision and semantic context, enhancing diagnostic accuracy. We validate our approach on the LIDC-IDRI dataset, employing standardized preprocessing techniques and balanced data splits. The proposed architecture integrates SwinUNETR for segmentation and a ResNet-based classifier with an attention-like mechanism that emphasizes clinically relevant regions. Experimental results demonstrate that MTL-LungNet achieves better segmentation performance than single-task baselines and provides malignancy prediction capabilities, emphasizing the promise of multi-task learning for robust, interpretable, and efficient lung cancer diagnosis.