Multimodal MRI Glioma IDH Genotyping Based On Deep Learning

Glioma is a highly prevalent primary malignant tumor of the central nervous system.In 2016,the World Health Organization included genotyping in the pathological classification of glioma.IDH genotyping is a critical factor in the diagnosis and treatment of these tumors.Pathological detection of glioma tissue samples is currently the gold standard for genotyping.However,invasive methods such as puncture or surgical resection are required to obtain these samples,which can cause neurological sequelae in patients and seriously affect their quality of life.Fortunately,the rapid development and widespread use of modern imaging technology has enabled non-invasive prediction and diagnosis of clinical diseases,including the possibility for accurate IDH genotyping of mesencephalic glioma patients.However,glioma MR images usually present with blurred boundaries and edge noise.In addition,there are differences in size,shape,and location of gliomas among different individuals.Therefore,these imaging characteristics pose significant challenges to IDH genotyping.The APT modality is a novel molecular imaging magnetic resonance technique.It reflects the APT signal intensity through the flow transfer of amide protons and water molecules.Unlike the commonly used T1CE modality in genotyping,APT modality is non-invasive and does not require patients to undergo contrast agent injections,which is a significant advantage in clinical practice.This paper focuses on utilizing the characteristics of glioma in APT modality images to build a classification model using deep learning and achieving IDH genotyping of glioma.The research is divided into two main parts:(1)The first part of the research involves combining the image characteristics of APT modalities and proposing a deep learning model based on a single modality.The effectiveness of APT is demonstrated by comparing the results of different modalities.Specifically,this study utilizes 3D APT image data to develop a Dual-Aware deep learning framework for IDH genotyping.To fully utilize APT image information,the Multi-scale Aware module is designed to extract multi-scale information,and the spatial attention mechanism is employed to fuse it and enhance global information for classification optimization.To address the lack of obvious edge contours in the tumor area for the APT modality,the Edge Aware module is designed to enhance edge features and further distinguish lesion and normal tissue areas,thereby improving the model’s sensitivity to lesion areas.(2)The second part of the research involves proposing a deep learning model based on multi-modality,leveraging the fusion characteristics of multi-modal information to further enhance the accuracy of IDH genotyping.Based on the features of APT modality images,this study employs the attention mechanism to extract and enhance deep features,enabling the fusion expression of APT modality and other modality information.This approach enhances the sensitivity of the model to the lesion area and improves classification performance.Experimental results demonstrate that this method not only achieves high genotyping accuracy but also exhibits excellent generalization ability.This paper employs a deep learning algorithm model constructed based on the image characteristics of the APT modality while deeply fusing multi-modal information to achieve IDH genotyping.Comparative experiments in the study verify the effectiveness of the APT modality in the identification task of glioma IDH genotyping.This approach is expected to fully utilize the non-invasive advantages of the APT modality in clinical practice,thereby avoiding the injection of contrast agents and achieving accurate identification of IDH genotyping.