Utilizing Deep Learning to Improve Diagnostic Accuracy in Glioma, Pituitary Tumors, and Meningiomas

BackgroundGlioma, pituitary tumors, and meningiomas constitute the major types of primary brain tumors. The challenge in achieving a definitive diagnosis stems from the brains complex structure, limited accessibility for precise imaging, and the resemblance between different types of tumors. An alternative and promising solution is the application of artificial intelligence (AI), specifically through deep learning models. MethodsThis study developed a brain tumor detection model using a comprehensive dataset of 7,023 MRI images. Included 1,621 images for glioma, 1,645 for meningioma, 1,757 for pituitary tumors, and 2,000 for non-tumor magnetic resonance imaging (MRI) images. We employed deep learning, specifically convolutional neural network (CNN). We utilized the ResNet-18 architecture, a widely used pre-trained model. We fine-tuned the pre-trained ResNet-18 on our dataset, which included MRI images classified into glioma, pituitary, and meningioma tumors. The fine-tuning was conducted over 10 epochs. After fine-tuning, the model that demonstrated the highest performance on validation data was selected as the best model ResultsThe deep learning model demonstrated exceptional performance across all diagnostic categories, with the training accuracy stabilizing at nearly 100% and validation accuracy closely mirroring this trend. The model achieved an AUC of 1.00 across all categories, with sensitivities of 95% for glioma, 98% for meningoma, 99% for pituitary tumors, and 100% for non-tumor MRI images. ConclusionOur study utilized a fine-tuned Residual Network 18 layers (ResNet-18) model in PyTorch to significantly enhance the diagnostic accuracy of primary brain tumors, specifically gliomas, meningiomas, and pituitary tumors. The model demonstrated high accuracy, indicating its potential to improve early diagnosis and patient outcomes. These findings suggest that integrating AI models like ResNet-18 into clinical workflows could lead to earlier detection and more precise prognostic evaluations, ultimately enhancing patient care and survival rates.