多模态超声-基因组指导甲状腺滤泡性肿瘤的诊断标志物研究

甲状腺滤泡癌侵袭性强、远处转移率高，早期、精准鉴别诊断滤泡性肿瘤（滤泡癌/滤泡腺瘤）将改善患者预后，术前鉴别诊断滤泡性肿瘤是目前临床亟需解决的问题。我们通过前期研究，分析公共数据库，发现PVALB、BCAN、RN7SK等差异基因，对鉴别诊断滤泡性肿瘤具有一定的价值，本项目拟在前期研究基础上，通过增加滤泡性肿瘤样本多样性，高通量测序验证差异基因及相关疾病通路，阐明滤泡癌侵袭性强的机制；并前瞻性收集滤泡性肿瘤多模态超声（灰阶、彩色、弹性超声及超声造影）图像，引用机器学习提出宏观影像学解读的可能，应用DeepFM算法完成模型训练，用LIME算法实现影像学特征可解释化；为明确宏观影像学特征与生物学性质的内在关系，建立图像特征与基因信息间的关联性，在宏、微观层面阐明：反映滤泡性肿瘤生物学行为的诊断标志物，为滤泡性肿瘤诊断难题提供新思路和研究基础。

Thyroid follicular carcinoma is highly aggressive and distant metastasis rate. Early and accurate differential diagnosis of follicular tumor (follicular carcinoma/ follicular adenoma) will improve the prognosis of patients. Preoperative differential diagnosis of follicular tumor is an urgent problem to be solved. We through the public databases research, found that PVALB, BCAN, RN7SK genes differential diagnosis of follicular tumor has a certain value, the project is proposed based on the previous research, by increasing the follicular tumor samples diversity, high-throughput sequencing validate genetic variations and related disease pathways, clarify follicular carcinoma invasive strong mechanism. This project prospectively collected multimodal ultrasonic images of thyroid follicular tumor (gray-scale ultrasound, color doppler ultrasound, elastography and contrast-enhanced ultrasound), extracted image features by machine learning, finished trained model by DeepFM algorithms and achieve the interpretation of imaging features by LIME algorithms. Further clarify the internal relationship between macroscopic imaging features and biological properties, establish the correlation between image features and genetic information. Elucidate the biological behavior of follicular tumor in the macro-level and micro-level. This project can provide new ideas and research basis for the diagnosis and treatment of thyroid follicular tumor.