Association of quantitative MRI-based radiomic features with prognostic factors and recurrence rate in oropharyngeal squamous cell carcinoma

Abstract

Radiomics focuses on extracting a large number of quantitative imaging features and testing both their correlation with clinical characteristics and their prognostic and predictive values. We propose a radiomic approach using magnetic resonance imaging (MRI) to decode tumor phenotype and local recurrence in oropharyngeal squamous cell carcinoma (OPSCC). The contrast-enhanced T1weighted sequences from baseline MRI examinations of OPSCC patients treated between 2008 and 2016 were retrospectively selected. Radiomic features were extracted using the IBEX software, and hierarchical clustering was applied to reduce features redundancy. The association of each radiomic feature with tumor grading and stage, HPV status, loco-regional recurrence within 2 years, considered as main endpoints, was assessed by univariate analysis and then corrected for multiple testing. Statistical analysis was performed with SAS/STAT® software. Thirty-two eligible cases were identified. For each patient, 1286 radiomic features were extracted, subsequently grouped into 16 clusters. Higher grading (G3 vs. G1/G2) was associated with lower 2 values of GOH/65Percentile and GOH/85Percentile features (p = 0.04 and 0.01, respectively). Positive HPV status was associated with higher values of GOH/10Percentile (p = 0.03) and lower values of GOH/90Percentile (p = 0.03). Loco-regional recurrence within 2 years was associated with higher values of GLCM3/4-7Correlation (p = 0.04) and lower values of GLCM3/2-1InformationMeasureCorr1 (p = 0.04). Results lost the statistical significance after correction for multiple testing. T stage was significantly correlated with 9 features, 4 of which (GLCM25/180-4InformationMeasureCorr2, Shape/MeanBreadth, GLCM25/90-1InverseDiffMomentNorm and GLCM3/6-1InformationMeasureCorr1) retained statistical significance after False Discovery Rate correction. MRI-based radiomics is a feasible and promising approach for the prediction of tumor phenotype and local recurrence in OPSCC. Some radiomic features seem to be correlated with tumor characteristics and oncologic outcome however, larger collaborative studies are warranted in order to increase the statistical power and to obtain robust and validated results.

MRI for Differentiation between HPV-Positive and HPV-Negative Oropharyngeal Squamous Cell Carcinoma: A Systematic Review

“…Twenty-one studies had a retrospective cohort study design [2,[19][20][21][22]25,26,28,29,31,32,35,[37][38][39]42,44,45,[47][48][49], and seven studies had a prospective cohort study design [23,27,40,41,43,46,50]. Of the twentyeight included papers, five papers focused on structural MRI sequences that were used in routine clinical examinations [19][20][21][22][23], nineteen focused on DWI sequences [2,25,[27][28][29]31,32,35,36,38,39,41,42,[44][45][46][47]...…”

“…Eleven studies used 1.5T MRI scanners [19,25,27,28,38,43,[45][46][47]49,50], nine studies used 3.0T MRI scanners [20,22,26,29,32,37,40,41,44], and eight studies used a combination of 1.5T and 3.0T MRI scanners [2,21,23,31,35,39,42,48]. Of the twenty-eight studies, nineteen investigated OPSCC exclusively [19][20][21][22][23]27,29,31,32,37,38,40,42,[44][45][46][47][48][49], three studies investigated oral cavity (OC)SCC as well as OPSCC [2,26,28], and ...…”

Radiomic Analysis for Human Papillomavirus Assessment in Oropharyngeal Carcinoma: Lessons and Pitfalls for the Next Future

“…MRI features have been widely investigated in many retrospective clinical studies with good performance results in assessing HPV+ status [31][32][33]. In this context Li et al [34] recently analyzed 141 patients and 2092 radiomic features extracted from both Primary Tumor (PT) and Largest Pathological Nodes (LN) on Contrast-Enhanced (CE) T1-weighted and T2-weighted MRI.…”

The Challenge of Choosing the Best Classification Method in Radiomic Analyses: Recommendations and Applications to Lung Cancer CT Images

“…The following are available online at https://www.mdpi.com/article/10 .3390/cancers13123088/s1, Supplementary Methods, Figure S1: Classification procedure flowchart, Figure S2: Correlation among radiomic features, Figure S3: Number of features correctly selected as associated with the outcome, Figure S4: Sensitivity and specificity for unbalanced datasets with and without SMOTE correction for large dataset, Figure S5: Sensitivity and specificity for unbalanced datasets with and without SMOTE correction for medium dataset, Figure S6: Sensitivity and specificity for unbalanced datasets with a without SMOTE correction for small dataset, Figure S7: Heatmap representing the predictive performance (AUC) in the validation set for unbalanced SMOTE samples, Table S1: Baseline characteristics of the original study population, Table S2: Wilcoxon pvalues for the association of features F1, F2 and F3 with outcome. References [29][30][31][32][49][50][51] are cited in the supplementary materials…”