Hybrid EfficientQNet for brain tumor detection using MRI images

Brain tumors (BT) result from the abnormal growth of brain cells and are associated with high morbidity and mortality rates. Malignant tumors spread quickly, while early-stage tumors grow slowly. Detection is challenging due to their varied sizes and shapes. To address this, EfficientQNet is proposed for effective BT detection using MRI. The process starts with preprocessing using the Fuzzy Local Information C-Means clustering model (FLICM) for Region of Interest (ROI) extraction and skull stripping, followed by SegNet for segmentation and image augmentation. Subsequently, texture features such as Correlation, Angular Second Moment, Inverse Difference Moment, Contrast, and Discrete Cosine Transform (DCT) with Fuzzy Local Binary Pattern (FLBP) are extracted. Finally, EfficientQNet is used for detection. Here, EfficientQNet combines the existing technologies, such as EfficientNet-B3-attn-2 with Deep Q-Learning to optimize layer configurations, achieving superior performance in brain tumor detection. Furthermore, EfficientQNet achieved an accuracy of 90.3 %, sensitivity of 93.2 %, specificity of 91.2 %, precision of 92.4 %, and F1-score of 92.8 %, with a loss of 9.7 %. The accuracy improvement over Fine-tuned Visual Geometry Group 16 (Fine-tuned VGG16), EfficientNet B0, Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), Ultra-Light Brain Tumor Detection (UL-BTD), Deep Learning-based Brain Tumor Detection and Classification using Magnetic Resonance Imaging (DLBTDC-MRI), and Parallel Deep Convolutional Neural Network (PDCNN) methods is 12.2 %, 9.63 %, 6.31 %, 5.42 %, 2.65 %, and 2.54 %, respectively.