Introduction: Brain tumors are abnormal accumulations of cells in the brain, categorized as cancerous or non-cancerous. The conventional diagnosis involves visually analyzing magnetic resonance images (MRI), a time-consuming and error-prone process. This study introduces an automated brain tumor diagnosis system that integrates a genetic algorithm (GA), artificial bee colony (ABC) algorithm, and convolutional neural network (CNN). These components facilitate pre-processing, feature extraction, feature selection, and tumor area segmentation. Additionally, the proposed system quantifies the brain tumor area. Notably, the system achieves a sensitivity of 0.9721 and specificity of 0.9743, indicating its potential superiority in comparison to recent research findings.
Methods: The automated brain tumor diagnosis system integrates a genetic algorithm (GA), artificial bee colony (ABC) algorithm, and convolutional neural network (CNN). The GA is employed for pre-processing, extracting high-quality features from the images. Subsequently, the ABC algorithm assists in selecting the most pertinent features within a short timeframe. The CNN is then employed to conduct comprehensive analysis, precluding the need for manual intervention. The system is calibrated and validated against a dataset comprising more than 3000 MRI images.
Results: The developed automated brain tumor diagnosis system showcases commendable performance. It effectively integrates the genetic algorithm, artificial bee colony algorithm, and convolutional neural network for accurate pre-processing, feature extraction, and segmentation of tumor areas. The system's accuracy is evidenced by achieving a sensitivity of 0.9721 and a specificity of 0.9743. These outcomes signify a significant advancement, aligning with or surpassing the outcomes of recent research efforts.
Conclusion: In conclusion, this study presents an innovative solution to the challenges associated with brain tumor diagnosis. By leveraging the synergy of genetic algorithm, artificial bee colony algorithm, and convolutional neural network, the proposed system achieves substantial automation in pre-processing, feature extraction, feature selection, and tumor area segmentation. The system's notable performance, with a sensitivity of 0.9721 and a specificity of 0.9743, underscores its potential to outperform existing methods. This research has implications for enhancing the accuracy and efficiency of brain tumor diagnosis, benefiting both medical professionals and patients.