Exploring the association between radiomics and gene expression in bladder cancer: a pilot study
Song, Yee Pei Elumalai, Thiraviyam Mistry, Hitesh Reeves, Kimberley Lane, Brian Dubec, Michael Jackson, S. West, Catharine M L Hoskin, Peter J Choudhury, Ananya
Song, Yee Pei
Elumalai, Thiraviyam
Mistry, Hitesh
Reeves, Kimberley
Lane, Brian
Dubec, Michael
Jackson, S.
West, Catharine M L
Hoskin, Peter J
Choudhury, Ananya
Citations
Altmetric:
Abstract
Purpose or Objective
A third of patients with bladder cancer present with advanced disease, requiring aggressive treatments. A curative
treatment option is trimodality bladder preserving therapy, comprising trans-urethral resection of bladder tumour (TURBT),
radical radiotherapy and radiosensitiser. Previous studies have shown that the West-24 gene signature is both prognostic
and predictive of survival outcomes following bladder preservation with hypoxia modification. Imaging biomarkers have the
potential for non-invasive assessment of tumour biology while allowing assessment of whole tumours. To date, there have
been no published studies on the association between radiomic biomarkers and gene expression in bladder cancer.
Materials and Methods
10 patients treated radically with concurrent chemoradiotherapy for histologically confirmed muscle-invasive bladder
cancer (MIBC) were included. Whole transcriptome sequencing using Clarion S transcriptomics was carried out on tumour
samples from pre-treatment TURBT. West-24 signature score and expression of individual genes making up the gene
signature were evaluated. MRI scans were carried out at 4 timepoints during treatment on 1.5T Aera Siemans scanner. 3D
T2 sequences were obtained. The tumour bed and whole bladder wall were contoured by two independent observers on
Raystation 7R. 93 features were extracted from the bladder wall and tumour bed using pyRadiomics. Next, unsupervised
feature selection removed features with low intraclass correlation coefficient (ICC) between the two observers and
between bladder wall and tumour. Unsupervised feature selection removed volume-correlated and redundant features,
using Spearman’s rank correlation coefficient. Remaining features were analysed with gene expression using Pearson
correlation. Due to small sample size, multiple hypothesis testing was not performed.
Results
19 of 93 features demonstrated a high ICC and remained after feature reduction. Levels of expression of 8 of these 19
features differed between tumour bed and bladder wall with no overlap of interquartile range. Expression of Gray Level Variance (GLV) had a strong correlations (R>0.7) with West-24 gene signature score (R=0.73) and expression of 4 individual
genes: COL5A1 (R=0.76), ITGA5 (R=0.73), TGFB1 (R=0.87) and PDLIM2 (0.73). The distribution of GLV was greater in tumour
bed than bladder wall (Fig 1). While Large Area High Gray Level Emphasis (LAHGLE) correlated with SLC16A1 expression
(R=0.71), the interquartile range of distribution of this feature in tumour bed and bladder wall overlaps and hence is
unlikely to reflect differences in tumour biology.Conclusion
This pilot study identified a potential radiomic feature as a surrogate for gene expression, thereby allowing a non-invasive
method of assessing for prognostic and predictive biomarkers. Further validation in a larger cohort would be worthwhile.
Description
Date
2021
Publisher
Collections
Keywords
Type
Meetings and Proceedings
Citation
Song YP, Elumalai T, Mistry H, Reeves K, Lane B, Dubec M, et al. Exploring the association between radiomics and gene expression in bladder cancer: a pilot study. Radiotherapy and Oncology. 2021;161:S1640-S1.