WolfCareNet: Optimized CNN-LSTM architecture using Grey Wolf Optimizer for early disease detection in smart...

Early detection of brain tumors greatly helps improve patient prognosis and allows for early clinical intervention. Until now, traditional ML methods such as decision trees, KNN, and fuzzy clustering have demonstrated average results in accuracy, scalability, and interpretability-less so when applied to medical imaging data like MRI scans. Many times, these methods will have limited feature representation and go through inappropriate optimization, leading to poor generalization when encountered with real-world clinical settings. The current research work puts forth WolfCareNet, an innovative CNN-LSTM model optimized with Grey Wolf Optimization (GWO). The model differs from traditional CNN-LSTM in that the GWO algorithm helps optimize the hyperparameters such as learning rate, batch size, dropout rate, and LSTM units, thus making the process more efficient. Temporal relationships from brain MRI images are detected using the LSTM technique, and the CNN technique detects strong spatial features. The GWO technique is used during the middle process to optimize the vital hyperparameters for the provision of optimal learning conditions. The model achieved an excellent output where the accuracy rate was 99.32%, the precision was 0.91, the recall was 0.93, F1-Score was 0.92, and the AUC-ROC was 0.97. In comparison with other existing models such as MLP, RNN, and fuzzy clustering, WolfCareNet performed better in terms of the performance of all the evaluation measures. The grad-CAM visualization technique also helped in interpreting the output by detecting tumors. These findings indicate the possible value of the proposed approach as a decision support system, although there is need for further research before its clinical application.