BTC-Net: A Boundary, Texture, and Context-Guided EfficientNetV۲B۰ Method for Multiclass Brain Tumor MRI Classification

Brain tumor classification from magnetic resonance imaging (MRI) plays an important role in computer-aided diagnosis and treatment planning. However, accurate classification remains challenging due to the significant intra-class variability and inter-class similarity among different tumor types. To address this issue, this study proposes BTC-Net, a Boundary–Texture–Context Guided EfficientNetV۲B۰ network for multiclass brain tumor classification. The proposed method utilizes EfficientNetV۲B۰ as a transfer learning backbone and extracts complementary features through three dedicated branches. A boundary branch is designed to emphasize structural and edge-related information, while a texture branch captures multi-scale local appearance patterns using parallel convolutions and channel attention. In addition, a CNN–Transformer context branch is introduced to model global semantic dependencies and contextual relationships within MRI images. The extracted features are integrated through an adaptive gated fusion mechanism to generate a discriminative representation for classification. Experiments were conducted on two publicly available brain MRI datasets. The Figshare dataset was evaluated using the official patient-wise five-fold cross-validation protocol. The proposed method achieved an average accuracy of ۹۴.۳۲%, precision of ۹۳.۴۵%, recall of ۹۳.۸۲%, and F۱-score of ۹۳.۵۳%. On the additional four-class brain MRI dataset, BTC-Net achieved an accuracy of ۹۸.۴۷% and an F۱-score of ۹۸.۳۷%, providing additional evidence of its performance in a different multiclass classification setting. Comparative analysis demonstrated that BTC-Net outperformed several recent brain tumor classification methods. The obtained results indicate that the joint exploitation of boundary, texture, and contextual information provides an effective solution for multiclass brain tumor MRI classification and may support future integration into intelligent digital and web-enabled healthcare information systems.

کلیدواژه ها:

Brain tumor ، MRI ، EfficientNetV۲B۰ Network ، Boundary Feature ، Texture feature ، Context Feature ، Transformer