CT-derived radiomics predict the growth rate of renal tumours in von Hippel-Lindau syndrome

S Singh ¹, F Dehghani Firouzabadi ¹, A Chaurasia ²

⁰Radiology and Imaging Sciences, Warren Grant Magnuson Clinical Center, National Institutes of Health, 10 Center Drive, Bethesda, MD 20892, USA.

Clinical Radiology +

|

February 21, 2024

View abstract on PubMed

Summary

Radiomic features can predict renal tumour growth in von Hippel-Lindau syndrome patients, aiding personalized surveillance. This helps improve patient outcomes by identifying fast-growing tumours early.

Area Of Science

Oncology
Radiology
Medical Imaging

Background

Von Hippel-Lindau (VHL) syndrome is a genetic disorder associated with an increased risk of developing various tumours, particularly renal cell carcinomas.
Effective surveillance strategies are crucial for managing VHL-associated renal tumours and improving patient prognosis.
Predicting tumour growth patterns can enable personalized surveillance plans.

Purpose Of The Study

To utilize radiomic features for predicting renal tumour growth patterns in patients with VHL syndrome.
To develop a model that assists in creating personalized surveillance strategies for VHL patients.
To ultimately improve patient outcomes through enhanced monitoring.

Main Methods

The study included 78 renal tumours from 55 patients with histopathologically confirmed clear cell renal cell carcinomas (ccRCCs).
Radiomic features were extracted from segmented tumours, and tumours were classified based on Volumetric Doubling Time (VDT) into fast-growing (<365 days) or slow-growing (≥365 days).
Machine learning models, including logistic regression and random forest classifiers, were employed to identify predictive features and assess model performance using ROC analysis.

Main Results

Fifty-five patients (mean age 42.2 ± 12.2 years) with 78 tumours were analyzed, with 25 classified as fast-growing and 53 as slow-growing.
The median volumetric and diametric growth rates were 1.71 cm³/year and 0.31 cm/year, respectively.
The most predictive radiomic features identified were wavelet-HLL_glcm_ldmn (AUC 0.80) and log-sigma-0-5-mm-3D_glszm_ZonePercentage (AUC 0.79). Random forest and logistic regression models showed AUCs of 0.73 and 0.74, respectively.

Conclusions

Radiomic features demonstrated a significant correlation with Volumetric Doubling Time (VDT).
These features can effectively predict the growth patterns of renal tumours in patients with VHL syndrome.
The findings support the use of radiomics in developing personalized surveillance plans for VHL patients, potentially leading to improved clinical outcomes.