Integrative radiomics expression predicts molecular subtypes of primary clear cell renal cell carcinoma. Issue 9 (September 2018)

• Record Type: Journal Article
• Authors: Yin, Q.; Hung, S.-C.; Rathmell, W.K.; Shen, L.; Wang, L.; Lin, W.; Fielding, J.R.; Khandani, A.H.; Woods, M.E.; Milowsky, M.I.; Brooks, S.A.; Wallen, E.M.; Shen, D.
• Is Part Of: Clinical radiology. Volume 73:Issue 9(2018)
• Page Start: 782
• Page End: 791
• Publication Date: 2018-09
• Subjects: Medical radiology -- Periodicals; Radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiology -- Periodicals; Societies, Medical -- Periodicals; Medical radiology; Radiotherapy; Electronic journals; Periodicals; 616.0757
• DOI: 10.1016/j.crad.2018.04.009
• Languages: English
• ISSNs: 0009-9260

Abstract

Abstract : Aim: To identify combined positron-emission tomography (PET)/magnetic resonance imaging (MRI)-based radiomics as a surrogate biomarker of intratumour disease risk for molecular subtype ccA and ccB in patients with primary clear cell renal cell carcinoma (ccRCC). Materials and methods: PET/MRI data were analysed retrospectively from eight patients. One hundred and sixty-eight radiomics features for each tumour sampling based on the regionally sampled tumours with 23 specimens were extracted. Sparse partial least squares discriminant analysis (SPLS-DA) was applied to feature screening on high-throughput radiomics features and project the selected features to low-dimensional intrinsic latent components as radiomics signatures. In addition, multilevel omics datasets were leveraged to explore the complementing information and elevate the discriminative ability. Results: The correct classification rate (CCR) for molecular subtype classification by SPLS-DA using only radiomics features was 86.96% with permutation test p = 7 × 10 − 4 . When multi-omics datasets including mRNA, microvascular density, and clinical parameters from each specimen were combined with radiomics features to refine the model of SPLS-DA, the best CCR was 95.65% with permutation test, p < 10 − 4 ; however, even in the case of generating the classification based on transcription features, which is the reference standard, there is roughly 10% classification ambiguity. Thus, this classification level (86.96–95.65%) of the proposed method represents the discriminating level that is consistent with reality. Conclusion: Featured with high accuracy, an integrated multi-omics model of PET/MRI-based radiomics could be the first non-invasive investigation for disease risk stratification and guidance of treatment in patients with primary ccRCC.