Accurate and reliable brain tumor identification is a critical component of cancer diagnosis and treatment planning. This task presents a significant challenge due to the extensive variability in tumor sizes, shapes, locations, scanning modalities, and acquisition protocols. Feature extraction is essential in brain tumor identification and has been improved by deep learning advancements, particularly CNNs. While state-of-the-art approaches predominantly employ parallel multiscale feature extraction to achieve high accuracy, they often suffer from computational inefficiency and limited generalization, especially when applied to small-scale datasets. This study primarily aims to develop a hybrid framework that integrates traditional machine learning and deep learning approaches to address challenges associated with small-scale image datasets. Specifically, a hierarchical feature representation is proposed for MRI brain tumor detection, leveraging advanced image preprocessing techniques. The framework incorporates deep convolutional neural networks, including customed CNNs, ResNet-18, and MobileNetV3, to extract discriminative features. The extracted features are then processed using two fuzzy ensemble methods, specifically the Gompertz and Mitscherlich functions to enhance the accuracy and robustness of tumor classification. The research highlights the potential of hierarchical synthesis and deep learning in enhancing diagnostic precision within medical imaging applications.