NephroNet: a calibration-aware, patient-disjoint benchmark for multiclass kidney CT classification with a compact depthwise-separable CNN
By
Yuxuan Dong
Masrufa Akter Muni
Rakibul Islam
Saima Tasnim
Sanjida Shahid Juthi
Md Jahirul Islam
Md AL Fassi
Md Sharif Robbani
Sufia Zareen
Yu Chen Hiu Lee
June 23, 2026
Clinical Scorecard: NephroNet: A Patient-Disjoint Benchmark for Multiclass Kidney CT Classification Utilizing a Calibration-Aware Depthwise-Separable CNN
At a Glance
Category Detail
Condition Renal Pathology Assessment Key Mechanisms Deep learning with a focus on probability calibration and patient-disjoint methodology. Target Population Patients undergoing renal CT imaging. Care Setting Multicenter clinical imaging environments.
Key Highlights
NephroNet achieves an accuracy of 0.9997 in renal CT classification. Utilizes a patient-disjoint, group-stratified hold-out benchmark. Incorporates calibration-aware reporting to improve probability assessments. Evaluates four classes: Normal, Cyst, Tumor, and Stone. Exceeds performance of budget-matched CNN and transformer baseline models. Guideline-Based Recommendations
Diagnosis
Utilize CT imaging for characterization of solid renal tumors and risk stratification of cystic renal lesions. Management
Implement a standardized protocol for renal CT classification to enhance reproducibility. Monitoring & Follow-up
Employ calibration metrics such as Brier score and Expected Calibration Error (ECE) for ongoing assessment. Risks
Address potential slice-level leakage and inadequate probability calibration in deep learning applications. Patient & Prescribing Data
Patients with renal pathology requiring imaging.
Focus on accurate classification to guide clinical decision-making.
Clinical Best Practices
Ensure patient-disjoint assessment to avoid optimistic bias. Combine discrimination metrics with calibration metrics for reliable outcomes. Utilize multicenter datasets for robust validation of imaging models. Related Resources & Content
