NephroNet: a calibration-aware, patient-disjoint benchmark for multiclass kidney CT classification with a compact depthwise-separable CNN - Takeaways - MDSpire

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

  • 0 min

Share

  • 1

    NephroNet is a depthwise-separable CNN designed for four-class renal CT classification using a 12,446-image multicenter dataset.

  • 2

    The model achieves an accuracy of 0.9997 and a macro-AUC of 0.9969 in the validation set, outperforming baseline models.

  • 3

    A patient-disjoint, group-stratified hold-out methodology is employed to prevent slice-level leakage in the evaluation process.

  • 4

    Calibration-aware reporting is integrated to ensure reliable probability outputs for clinical decision-making.

  • 5

    All findings are derived from a single-center location, necessitating external validation for broader applicability.

Original Source(s)

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

Related Content

In-vitro flow assessment study of intra-saccular endovascular devices for brain aneurysm treatment: SEAL™ vs. WEB™

Compensatory activation of the deltoid and remaining rotator cuff muscles in patients with rotator cuff tears using real-time tissue elastography

Pose-aware deep perceptual similarity for iterative 2D/3D registration of knee joints using contrastive learning