Image-based deep learning model to predict stoma-site incisional hernia in patients with temporary ileostomy: A retrospective study

Zhongyi Dong; Jianhua Cai; Haigang Geng; Bo Ni; Mengqing Yuan; Yeqian Zhang; Xiang Xia; Haoyu Zhang; Jie Zhang; Chunchao Zhu; Un Wai Choi; Aksara Regmi; Cheok I Chan; Cara Kou Yan; Yan Gu; Hui Cao; Zizhen Zhang

doi:10.1016/j.isci.2024.111235

Image-based deep learning model to predict stoma-site incisional hernia in patients with temporary ileostomy: A retrospective study

iScience. 2024 Oct 22;27(11):111235. doi: 10.1016/j.isci.2024.111235. eCollection 2024 Nov 15.

Authors

Zhongyi Dong¹, Jianhua Cai², Haigang Geng¹, Bo Ni¹, Mengqing Yuan³, Yeqian Zhang¹, Xiang Xia¹, Haoyu Zhang¹, Jie Zhang⁴, Chunchao Zhu¹, Un Wai Choi⁵, Aksara Regmi⁵, Cheok I Chan⁵, Cara Kou Yan⁵, Yan Gu², Hui Cao¹, Zizhen Zhang¹

Affiliations

¹ Department of Gastrointestinal Surgery, Renji Hospital, School of Medicine, Shanghai Jiaotong University, Shanghai 200127, P.R. China.
² Department of General Surgery, Fudan University Affiliated Huadong Hospital, Shanghai 200040, P.R. China.
³ School of Science, The Hongkong University of Science and Technology, Hongkong 999077, P.R. China.
⁴ Department of Interventional Oncology, Renji Hospital, School of Medicine, Shanghai Jiaotong University, Shanghai 200127, P.R. China.
⁵ School of Medicine, Shanghai Jiaotong University, Shanghai 200025, P.R. China.

Abstract

The prophylactic implantation of biological mesh can effectively prevent the occurrence of stoma-site incisional hernia (SSIH) in patients undergoing stoma retraction. Therefore, our study prospectively established and validated a mixed model, which combined radiomics, stepwise regression, and deep learning for the prediction of SSIH in patients with temporary ileostomy. The mixed model showed good discrimination of the SSIH patients on all cohorts, which outperformed deep learning, radiomics, and clinical models alone (overall area under the curve [AUC]: 0.947 in the primary cohort, 0.876 in the external validation cohort 1, and 0.776 in the external validation cohort 2). Moreover, the sensitivity, specificity, and precision for predicting SSIH were improved in the mixed model. Thus, the mixed model can provide more information for SSIH precaution and clinical decision-making.

Keywords: Computing methodology; Health sciences; Medical imaging.