基于MRI影像基因组学预测垂体腺瘤侵袭的人工智能模型建立与临床评估

垂体腺瘤是最常见的颅内原发肿瘤之一，约半数的垂体腺瘤常侵袭周围重要结构，造成手术全切困难、易复发。项目组建立了垂体腺瘤生物样本库、基因组学数据库和影像组学数据库，发现了11个与侵袭相关的分子标志物，初步完成了基于3个影像组学特征的侵袭性评估模型。本课题拟围绕“基于MRI影像基因组学预测垂体腺瘤侵袭的人工智能模型建立与临床评估”这一关键问题，在既往单纯影像组学评估模型基础上，完善对应样本的转录组学数据，应用CNN、LASSO、SVM等机器学习与人工智能方法，建立垂体腺瘤侵袭性评估模型。在前瞻性临床队列中，完成MRI影像特征提取及转录组测序，通过bootstrap、ROC曲线等方法，评价模型的准确度、敏感性、特异性。通过影像组学与转录组数据的人工智能关联分析，进一步明确模型的生物学意义。本项目有望建立更精准、更智能的垂体腺瘤侵袭性评估模型，为开展临床应用及干预奠定基础。

Pituitary adenoma is one of the most common primary intracranial tumors. About half of the tumors often invade important structures around it, making surgical resection difficult and prone to recurrence. We have already established a bio-bank of pituitary adenoma samples, a genomics database, and a radiomics database. Based on the databases, we have found 11 invasion associated markers, and established a primary invasion prediction model based on 3 radiomic features. This project intends to focus on "the establishment and clinical evaluation of an artificial intelligence model for predicting pituitary adenoma invasion based on MRI radiogenomics". Based on the previous radiomics evaluation model, the transcriptomics data of the corresponding samples will be added to improve the model. CNN, LASSO, SVM and other machine learning and artificial intelligence methods will be used to establish an evaluation model for pituitary adenoma invasiveness prediction. We will extract MRI features and complete RNA sequencing and the radiogenomic data will be used to evaluate the accuracy, sensitivity, and specificity of the model by methods such as bootstrap and ROC curves. The biological significance of the model will be further clarified by the artificial intelligence correlation analysis of radiomics and transcriptome data. This project is expected to establish a more accurate and intelligent invasion evaluation model for pituitary adenomas, laying a foundation for clinical application and intervention.