Artificial Intelligence for the prediction of postoperative/postablative renal function in elderly and/or comorbid patients

• Applicant: Privatdozentin Dr. Annemarie Uhlig
• Subject Area: Reproductive Medicine, Urology ; Nuclear Medicine, Radiotherapy, Radiobiology
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 511948726

Acute Kidney Injury (AKI) and Chronic Kidney Disease (CKD) are serious complications of the surgical/ablative therapy of localized kidney cancer. Old and comorbid patients are at increased risk for posttherapeutic AKI-/CKD. In particular, these patients suffer from kidney injury, with relevant impact on “in-hospital” mortality rates and short- and long- term patient survival.Currently, there are only few prediction models for posttherappeutic AKI/CKD, and the existing models are of low clinical relevance. Although initial studies show promising correlations between cross-sectional pretherapeutic imaging and posttherapeutic kidney function, current AKI/CKD prediction models do not incorporate imaging data. Therefore, the superior aim of this research proposal is the development of risk models for AKI/CKD which combine clinical data and pretherapeutic computed tomography (CT) using artificial intelligence (AI). This project focusses on the evaluation of old and comorbid kidney cancer patients to provide optimized and individualized therapeutic concepts for this vulnerable patient population. The 12-month retrospective cohort study uses 2 AI approaches for evaluation of clinical data and information from CT imaging studies: 1) Use of “Radiomics” in combination with machine leaning (ML) 2) Use of Deep Leaning (DL) algorithms. The following comparisons will be made: i) Prediction of AKI/CKD based on clinical data alone ii) Prediction of AKI/CKD based on clinical data + Radiomics and ML iii) Prediction of AKI/CKD based on clinical data + DL The approach with the best predictive performance will be used for further analyses. The comparisons of patients <65 years of age vs. ≥65 years, and of patients without or with only few comorbidities vs. comorbid patients (Charlson Comorbidity Index <3 vs. ≥3) are of particular relevance for the project: This study will culminate in risk models for AKI and CKD, with separate models for age and comorbidity groups if deemed useful. The project comprises 6 work packages, of which the first completes an already established clinical-radiologic database. Student research assistants and a data manager will be working on this milestone. The development of prediction algorithms and their validation will be completed under the leadership of the applicant. The analyses using Radiomics an ML will be performed in cooperation with Dr. Andreas Leha (Institute for medical Biostatistics, University Medical Center Göttingen, UMG) and with Mrs. Hazal Timuçin (Institute for medical Bioinformatics, UMG) for DL. All further analyses (comparison of the performance of the algorithms, development of age/comorbidity-dependent risk scores), and summary of the results will be completed by the applicant.

• DFG Programme: Research Grants