Abstract: FR-PO011
Histopathological Prediction of CT-Based Radiomic Imaging Biomarkers in Native Kidney Biopsies
Session Information
- AI, Digital Health, Data Science - II
November 03, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Kim, Ji Eun, Inha University Hospital, Incheon, Korea (the Republic of)
- Kim, Kipyo, Inha University Hospital, Incheon, Korea (the Republic of)
Background
In recent years, there has been a growing interest in radiomics as a quantitative approach for image analysis. This study aims to explore the relationship between radiomic features extracted from CT scans and histological findings obtained from kidney biopsies.
Methods
We retrospectively enrolled participants who underwent abdomen CT scan within 7 days before native kidney biopsy. Three-dimensional kidney segmentation was performed in two different methods; the entire kidney parenchyma and isovolumumic cortical area. We extracted various radiomics features such as shape, first-order, and texture features from the CT images. The histological findings were assessed using a semiquantitative scoring system, which assigned severity grades to various parameters including interstitial fibrosis (IF), tubular atrophy (TA), glomerulosclerosis (GS), interstitial inflammation (II), and arterial intimal thickening (IT).
Results
Total of 124 patients were included in the main analysis. The ROC curve analysis of the extracted radiomic features revealed higher AUC values for moderate IF, TA, and II in both the total kidney parenchyma and cortex. However, the ability to distinguish GS and IT was relatively lower. The most discriminatory features extracted from total kidney for IF and TA were GLCM IDMN (LLH-wavelet) and GLCM IDN (LLH-wavelet), with the AUCs approximately 0.83. For the kidney cortex, GLRLM GLNU (HLL-wavelet), GLDM DE (HLL-wavelet), and GLRLM RV (HLH-wavelet) had the texture features with the highest AUCs for IF and TA, with the AUCs of 0.84-0.88. Texture features extracted from kidney cortex showed higher correlation with histologic features. We found that CT-based texture features consist of volume-dependent and independent components. Volume-independent texture features extracted from kidney cortex showed more a higher degree of correlation with chronic histologic scores.
Conclusion
In conclusion, our findings suggested the potential of CT-based radiomics in predicting chronic histological findings in kidney biopsies.