Deep learning technology enables the transformation of healthcare, focusing on the deployment of lightweight or hybrid diagnostic models for insufficiently resourced medical facilities. Early-stage brain tumor diagnosis supports doctors in developing better treatment strategies and enhancing prognosis results. In this work, we present a CNN-based model for automated brain tumor identification through magnetic resonance imaging (MRI). Our model achieved better generalization capability through pre-processed imaging that involved normalization followed by segmentation. Optimization was carried out using hyperparameter tuning, early stopping and model checkpointing. The model exhibited remarkable performance, attaining 99% accuracy, sensitivity and specificity and precision. Reliability and robustness measurements were validated through comprehensive evaluation metrics, including confusion matrix analysis, receiver operating characteristic (ROC) curve and F1-score. The findings indicate the effectiveness of the proposed CNN framework as an effective and computationally feasible decision-support system in aid of radiologists in the early detection of brain tumors, especially in the context of limited resources healthcare settings.
Reddy et al. (Fri,) studied this question.