<p>This study proposes a comprehensive and clinically oriented framework for the automatic segmentation of FLAIR hyperintense lesions and classification of gliomas into low-grade (LGG) and high-grade (HGG) categories by integrating deep learning (DL) based segmentation with machine learning (ML) based classification using quantitative DCE-MRI parameters. While several studies have explored hybrid DL–ML approaches for glioma analysis, the proposed framework combines automated segmentation using optimized DL architectures with classification based on physiologically meaningful perfusion parameters, providing a clinically interpretable pipeline. Three DL architectures, U-Net, MultiRes U-Net, and nnU-Net were systematically evaluated on the BraTS 2021 challenge dataset (1251 glioma patients) and a local hospital dataset comprising 130 glioma patients (grade 2: 43, grade 3: 21, grade 4: 66) for tumor segmentation task. The optimized model based on U-Net architecture provided high accuracy in segmenting FLAIR hyperintense lesions, with a mean dice similarity coefficient of 0.84 on the local hospital test dataset. For classifying gliomas, multiple ML classifiers were assessed using quantitative DCE-MRI features, with the support vector machine (RBF kernel) achieving a classification accuracy of 87% on the local test dataset. Overall, the proposed DL-ML pipeline demonstrates accurate, computationally efficient, and generalizable performance for glioma segmentation and grading, highlighting its potential utility as a decision-support tool for radiologists in clinical brain tumor assessment and treatment planning.</p>