Application of Feature Re‐Calibration Network Combined With Transformer in MRI Brain Tumour Diagnosis

Early and accurate diagnosis of brain tumours is crucial for developing appropriate treatment plans and improving survival rates. With the development of Smart Healthcare, the complete reliance on manual analysis of MRI images is becoming smarter. To assist doctors in diagnosing brain tumour diseases more efficiently, a feature re‐calibration network (FRCNet)is proposed in this paper. First, a partially transformer block (PTB), a lightweight CNN‐Transformer dual‐branch parallel feature extraction module, was designed to solve the difficult brain tumour classification problem. Second, for the fuzzy characteristics of brain tumour boundaries, re‐calibrated feature pyramid network (RFPN) was constructed to recalibrate the tumour location with fine‐grained tumour contours to improve the detection accuracy of brain tumours. Finally, the improved wise‐MPDIoU loss function is utilized to improve the model's convergence speed and regression accuracy. Experiments show that FRCNet achieves 99.5% classification accuracy and 99.3% mean average precision for the four classification tasks of brain tumours while maintaining high detection speed. The number of parameters of the model is only 3.49 M, which is better than other advanced algorithms in terms of comprehensive performance and achieves a balance between accuracy and lightweight. This provides an intelligent new method for brain tumour detection and classification.