Accurate predictions of clinical stage in lung cancer is vital for early intervention and effective treatment planning. In this study, we propose an attention-based deep learning framework leveraging multimodal data from cBioPortal repository to predict clinical stage of lung cancer patients. After preprocessing, class balancing and normalization, exploratory data analysis (EDA) was performed to assess the patterns in the data. The proposed model trained with adam optimizer, achieved high classification accuracy (98.43%) on independent test set and demonstrated robust performance across all clinical stages with precision, recall and F1-score approaching 1.00. The training and Validation curves confirmed stable convergence without overfitting. To enhance model transparency local interpretability model agnostic explanations (LIME) were wemployed, identifying influential features. We further performed computational complexity analysis demonstrating that attention-based model sacles linearly with data volume, maintaining efficient training time even on large dataset, supporting its suitability for real world medical applications. These results confirm the effectiveness and interpretability of proposed framework and its potential applications in clinical decision support for personalized oncology.
Kayani et al. (Sat,) studied this question.