ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Abstract: SA-PO0008

Artificial Intelligence (AI)-Driven Estimation of Total Kidney Volume (TKV) in ADPKD

Session Information

Category: Artificial Intelligence, Digital Health, and Data Science

  • 300 Artificial Intelligence, Digital Health, and Data Science

Authors

  • Yousef Kalafi, Elham, Perceptive Informatics UK Limited, Burlington, Massachusetts, United States
  • Kadumberi, Anand Preshob, Perceptive Informatics UK Limited, Burlington, Massachusetts, United States
  • Michalski, Andrew S, Perceptive Informatics UK Limited, Burlington, Massachusetts, United States
  • Sood, Rohit, Perceptive Informatics UK Limited, Burlington, Massachusetts, United States
Background

ADPKD is characterized by renal cysts and bilateral renal enlargement. The US FDA has provided guidance for imaging based TKV assessment as qualified prognostic biomarker in clinical trials. AI driven methods allow automated kidney segmentation, high repeatability and reproducibility. Objective: To develop an AI-driven methodology to estimate TKV on MRI.

Methods

Retrospective cohort of MRIs from CRISP ADPKD study was analyzed (T1w, T2w). One reader (15 yrs experience) annotated the T2w images. Data prep - Rigid registeration of T1w to matching T2w, splitting the dataset into training, validation and test sets.
Model Training- Models used the nnU-Netv2 with self-tunable 3D U-Net that adjusts hyperparameters based on training data. Three baselines were trained- T1w model with 43 preprocessed T1w image sets(ISs),T2w model with 67 T2w ISs, multi-modal model using channel-wise concatenated 66 registered T1w and T2w ISs. Models were trained with manual kidney segmentations and tested on a held-out test set (24 ISs).
Additional experiments involved pretraining on public MRI (AMOS,CHAOS,Totalsegmentator,CKD) and CT (AbdomenCT-1K) datasets, followed by fine-tuning on the T2w images.
Evaluation Metrics: The performance of all trained models was evaluated on the held-out test set using Dice Similarity Coefficient (DSC),Volume Error (VE), Hausdorff Distance (HD) and Intersection over union (IoU).

Results

Of all baseline models, T2w model performed the best in terms of all metrics, with the highest median DSC 0.94, IoU 0.88, HD 13.9 and VE 57.5cc (Fig 1a). The T2w fine tuned model demonstrated comparable results and superior generalizability, particularly in cases with poor image quality. Models built with T1w underperformed compared to those on T2w images alone. Fig 1b and 1c shows high (DSC 0.96) and low accuracy (DSC 0.68) examples.

Conclusion

T2w images are sufficient and optimal for kidney segmentation in ADPKD. Pretraining on other pathology and/or healthy MRI datasets enhances generalizability of model performance. Absence of T1w annotations likely caused the poor performance of T1w models and will be explored in future work.

Funding

  • NIDDK Support

Digital Object Identifier (DOI)