Contribution and Clinical Significance of 3D and 2D Deep Learning and Radimoics Features Based on MR-T1W Images in Joint Diagnosis of Parkinson's Disease
Yan Mo,Xiaojun Guan,Haoyi Xiong,Tao Guo,Cheng Zhou,Jingjing Wu,Xueqin Bai,Jianmei Qin,Haoting Wu,Xiaocao Liu,Jiaqi Wen,Jingwen Chen,Qingzhong Wang,Xuhong Li,Jiang Bian,Xiaojun Xu,Dejing Dou
DOI: https://doi.org/10.2139/ssrn.4053479
2022-01-01
SSRN Electronic Journal
Abstract:Purpose: To develop and evaluate the diagnostic ability of MRI image deep features and radiomics features between normal controls and Parkinson's disease (PD).Methods: MRI T1-weighted images diagnosed with PD from four Centers were collected, with sample sizes of 401,421,43 and 40. Clinical factors including age, gender, U3, H-Y stage and MMSE were statistically analyzed. 2D deep, 3D deep and radiomics features of MRI images were extracted by 2D-Resnet50, 3D-Resnet50 and Pyradiomic respectively. The classification models were trained and tested internally in Center 1 and tested externally in Center 2,3,4. The correlation between clinical and image features was analyzed, and CAM was used to explain the activation area of deep learning and visualized. Finally, independent factors were obtained from the above features to construct a Nomogram, and features’ contributions were quantified, and the AUCs, calibration and DCA curves were obtained.Results: There were significant differences between PD and NC in five clinical factors(p <0.05). The AUCs of 2D deep and 3D deep features in internal training set (Centre1) were 0.9062 and 0.8426, in internal test set (Centre1) were 0.8096 and 0.6880, well in external test set (Centre2: 0.7948 and 0.7649, Centre3: 0.7292 and 0.6111, Centre4: 0.7525 and 0.6075) . LR-radiomics classifier’s AUC performed well in internal training ( 0.8651) and internal test set (0.8335), and verified satisfactory AUC in three external test sets (Centre2: 0.7885, Centre3:0.6181, Centre4:0.74). A Nomogram was constructed and has the best AUC in internal training set (0.9437) and internal test set (0.8501), and was significantly higher than other single feature models (P <0.05), and showed the highest F1-score and the lowest FPR and FNR.Conclusion: This study demonstrated the effectiveness of brain MRI image features in rapid adjunctive diagnosis of PD and had the potential to contribute to early control of PD.Funding Information: This work was supported by the National Natural Science Foundation of China (Grant Nos. 82171888, 82001767, and 81971577), the Natural Science Foundation of Zhejiang Province (Grant No. LY22H180002 and LQ21H180008), the China Postdoctoral Science Foundation (Grant Nos. 2021T140599 and 2019M662082), and the 13th Five-year Plan for National Key Research and Development Program of China (Grant No. 2016YFC1306600). Declaration of Interests: All authors declare no competing interests.Ethics Approval Statement: All patients with PD and normal controls signed informed consent forms in accordance with the approval of the Medical Ethic Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine (Nos.2012-26 and 2017-008).