Brain Tumor Detection in MRI Images Using Artificial Intelligence with a Transfer Learning Approach and Decision Threshold Optimization

Timely and accurate detection of brain tumors in MRI images is one of the most critical challenges in diagnostic radiology. This study develops a binary image classification model (tumor/healthy) using the lightweight and efficient MobileNetV2 architecture combined with a transfer learning strategy. Considering the clinical importance of minimizing false negatives (FN), two key innovations were introduced to enhance diagnostic safety: first, applying asymmetric class weighting (a weight of 1.5 for the tumor class) during the training phase to impose a stronger penalty on FN errors; and second, optimizing the decision threshold from 0.5 to 0.3 during the evaluation phase. The final evaluation on the test set (203 images) demonstrated the effectiveness of this approach: at the 0.3 threshold, the model achieved 100% recall for the tumor class, indicating a complete elimination of false negatives (FN=0). This result ensures the highest level of sensitivity and diagnostic safety required for computer-aided detection (CAD) systems in medical screening. In essence, this research proposes an innovative and efficient approach for developing a model capable of faster and more accurate detection of brain tumors.