The CT-based Deep Learning Model Predicts Complications in Partial Nephrectomy

The CT-based Deep Learning Model Outperforms Traditional Anatomical Classification Models in Preoperatively Predicting Complications and Risk Grade in Partial Nephrectomy

The investigators combine radiomics and deep learning to analyze the lesions more thoroughly, aiming for a more accurate prediction of complications in partial nephrectomy, and compare this approach with traditional models.

Study Overview

• Status: Completed
• Conditions: Renal Cell Carcinoma (RCC); Renal Cyst

Detailed Description

In this study, patients diagnosed with renal cell carcinoma or renal cyst who underwent partial nephrectomy across multiple centers was included. And the participants were excluded if they had (a) missing or unavailable imaging data or (b) no available enhanced CT images. The cohort was divided into training and test sets at a 7:3 ratio. After that, the radiomics features were extracted from the images, and lasso regression was used to select features. Then a deep learning model was developed to predict complications and risk grades and compared with traditional classification models (RENAL and PADUA), demonstrating superior applicability.

• Study Type: Observational
• Enrollment (Actual): 1474

Contacts and Locations

Study Locations

• China
  • Xuhui District
    • Shanghai, Xuhui District, China, 200032
      • Name: Zhongshan Hospital Fudan University, Location: 180th Fenglin Road, Xuhui District, Shanghai, China

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Child; Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

The study population includes patients diagnosed with renal cell carcinoma or renal cyst who underwent partial nephrectomy in the participated centers. Clinical and imaging data were retrospectively collected from medical records, including demographic characteristics (age, gender, BMI), tumor location (left or right kidney), surgical details (surgical approach, ischemia time), and perioperative complications.

Patients were included based on the availability of complete clinical, surgical, and imaging data. Exclusion criteria comprised individuals with missing or unavailable imaging data, or no available enhanced CT images. The study aims to combine CT-based radiomics features and clinical features to develop a deep learning model to predict perioperative complications of partial nephrectomy, and compare with traditional anatomical classification models.

Description

Inclusion Criteria:

• Clinical diagnosis of renal cell carcinoma or renal cyst
• Underwent partial nephrectomy between June 2014 and July 2024

Exclusion Criteria:

• Missing or unavailable imaging data
• No available enhanced CT images

Study Plan

How is the study designed?

Number of groups / cohorts

2

Cohorts and Interventions

Group / Cohort
Complication 1; Patients who experienced perioperative complications during the partial nephrectomy
Complication 0; Patients who didn't experience perioperative complications during the partial nephrectomy

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
whether complications occurred	Retrospectively review the medical record system to determine whether patients developed postoperative complications.	perioperatively

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Patients' risk grade	Based on the widely recognized Clavien-Dindo classification (CDC) system for surgical complications, these complications were categorized into four grades: Grade I, II, III, and IV. Risk grade was assigned accordingly: "no risk" is defined as no complications occurred, "grade low" is defined as the highest level of complication being Grade I, "grade moderate" is defined as the highest level of complication being Grade II, and "grade high" is defined as complications of Grade III or higher, which are life-threatening.	perioperatively

Collaborators and Investigators

• Sponsor: Du Lingzhi
• Collaborators: Shanghai Zhongshan Hospital; Xuhui Central Hospital, Shanghai; Minhang Hospital, Fudan University

Study record dates

Study Major Dates

• Study Start (Actual): June 1, 2024
• Primary Completion (Actual): December 31, 2024
• Study Completion (Actual): February 28, 2025

Study Registration Dates

• First Submitted: March 8, 2025
• First Submitted That Met QC Criteria: March 12, 2025
• First Posted (Actual): March 25, 2025

Study Record Updates

• Last Update Posted (Actual): March 25, 2025
• Last Update Submitted That Met QC Criteria: March 12, 2025
• Last Verified: March 1, 2025

More Information

Terms related to this study

• Keywords: partial nephrectomy; CT-based deep learning; complication prediction; traditional classification model
• Additional Relevant MeSH Terms: Urogenital Diseases; Urogenital Neoplasms; Neoplasms by Site; Neoplasms; Male Urogenital Diseases; Kidney Diseases; Urologic Diseases; Female Urogenital Diseases; Female Urogenital Diseases and Pregnancy Complications; Neoplasms by Histologic Type; Neoplasms, Glandular and Epithelial; Adenocarcinoma; Urologic Neoplasms; Carcinoma; Kidney Neoplasms; Carcinoma, Renal Cell
• Other Study ID Numbers: zsurologyDLMforPNcomplication

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: YES
• IPD Plan Description: Clinical data and extracted radiomics feature data, excluding patient information.
• IPD Sharing Time Frame: Within six months after publication in the journal.
• IPD Sharing Access Criteria: The data supporting this study are available from the enrolled institutions, but restrictions apply to their availability due to privacy reasons. Data can be accessed upon reasonable request from the corresponding author.
• IPD Sharing Supporting Information Type: STUDY_PROTOCOL

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No