ABSTRACT Gliomas are very common malignant brain lesions with distinct grades that relatively estimate the survival rate in affected persons. Tumor grading and segmentation are significant and common for treatment planning and diagnosis. The approach of glioma grading is highly performed by the imaging models at clinical practice. The accurate glioma grading in the operation time supports the clinicians in estimating the highest excision region. But, it is highly complex for medical experts to estimate the glioma grades alone at the time of operation. Hence, an accurate and rapid tumor tissue diagnosis during the surgery period is essential. Nowadays, deep learning strategies have been utilized for the evaluation of medical images and the diagnosis of diseases. But, the problems of near‐infrared images such as low resolution, noise, and small scale cause the complexity of training a promising model. Multimodal imaging can offer harmonizing data to improve the system performance, yet continuously developing a proper model and employing the multimodal data is complex. Therefore, this work proposes a glioma grading classification approach using multimodal medical images and deep learning. Initially, the necessary multimodal images such as CT and MRI are accumulated from benchmark datasets, and the garnered multimodal images are fused with the help of discrete wavelet transformation (DWT). Further, the fused multimodal image is fed into the tumor segmentation module, where the multilevel attention‐based transformer residual Unet++ (MA‐TRUnet++) technique is supported. Furthermore, the tumor‐segmented images are subjected to the adaptive dilated residual network (ADRNet) module. In this module, at first, the brain tumor is detected and classified as normal and abnormal. If the tumor is malignant, then the same ADRNet technique is employed for classifying the glioma grades. Here, the revised chameleon swarm algorithm (RCSA) is utilized for fine tuning the parameters in ADRNet. Finally, the classified glioma grade is achieved and the performance analysis is conducted for the presented work by comparing it with the baseline approaches to ensure the efficacy of the offered model. The github link of the developed fusion‐based glioma detection model is given in https://github.com/gurusigaamani0/Multimodel‐Medical‐Image‐Fusion/tree/main .
Muthulingam et al. (Sun,) studied this question.