Multi-centre radiomics for prediction of recurrence following radical radiotherapy for head and neck cancers: Consequences of feature selection, machine learning classifiers and batch-effect harmonization

Background and purpose

Radiomics models trained with limited single institution data are often not reproducible and generalisable. We developed radiomics models that predict loco-regional recurrence within two years of radiotherapy with private and public datasets and their combinations, to simulate small and multi-institutional studies and study the responsiveness of the models to feature selection, machine learning algorithms, centre-effect harmonization and increased dataset sizes.

Materials and methods

562 patients histologically confirmed and treated for locally advanced head-and-neck cancer (LA-HNC) from two public and two private datasets; one private dataset exclusively reserved for validation. Clinical contours of primary tumours were not recontoured and were used for Pyradiomics based feature extraction. ComBat harmonization was applied, and LASSO-Logistic Regression (LR) and Support Vector Machine (SVM) models were built. 95% confidence interval (CI) of 1000 bootstrapped area-under-the-Receiver-operating-curves (AUC) provided predictive performance. Responsiveness of the models' performance to the choice of feature selection methods, ComBat harmonization, machine learning classifier, single and pooled data was evaluated.

Results

LASSO and SelectKBest selected 14 and 16 features, respectively; three were overlapping. Without ComBat, the LR and SVM models for three institutional data showed AUCs (CI) of 0.513 (0.481-0.559) and 0.632 (0.586-0.665), respectively. Performances following ComBat revealed AUCs of 0.559 (0.536-0.590) and 0.662 (0.606-0.690), respectively. Compared to single cohort AUCs (0.562-0.629), SVM models from pooled data performed significantly better at AUC = 0.680.

Conclusions

Multi-institutional retrospective data accentuates the existing variabilities that affect radiomics. Carefully designed prospective, multi-institutional studies and data sharing are necessary for clinically relevant head-and-neck cancer prognostication models.

Abstract P1-06-10: Next generation sequencing (NGS) reveals high mutation rates in all established breast cancer subtypes with subtype-specific patterns

Introduction. Next-generation sequencing (NGS) technologies offer the possibility for assessment of multiple genes for their somatic mutations and may elucidate the driver genetic variations involved in carcinogenesis and disease progression. Among the various subtypes of breast cancer (BC), the triple-negative subgroup (TNBC) is characterized by poor prognosis and a lack of reliable tumor markers when compared to the luminal-A/-B and HER-2 positive subtypes. The aim of this study was to characterize genetic variations in primary BC obtained from a cohort of 159 Norwegian patients using a NGS panel consisting of 44 BC relevant genes. Our goal was to compare the genetic variations between BC subtypes in general with special emphasis on the TNBC subtype (over 40% of the patients).

Methods. Genomic DNA was extracted from paraffin embedded formalin fixed (FFPE) tissue obtained from 160 consecutive patients diagnosed with a primary BC at our hospital. The DNA samples were analyzed by next-generation sequencing (NGS) using Human Breast Cancer GeneRead DNAseq Targeted Panel V2 (Qiagen). The panel consists of a collection of PCR primers for targeted enrichment of the coding region of 44 genes commonly mutated in BC. Target enrichment and library construction was performed according to the GeneReader workflow (Qiagen) and paired end sequencing was performed on a NextSeq 500 sequencer (Illumina) running 2 x 150 bp chemistry Version 2. Data analysis including alignment to the reference genome hg19 and variant calling was performed using Qiagen's online Ingenuity Variant analysis.

Results. The ingenuity variant analysis classified the genetic mutations according to their clinical significance into four groups: pathogenic, likely pathogenic, benign and likely benign. We present only mutations in genes that are characterized as pathogenic or likely pathogenic, where the term "likely pathogenic" indicates greater than 90% certainty of the mutation being pathogenic (as defined by the American College of Medical Genetics and Genomics). Genetic variations were mostly observed in a subset of 44 genes included in the breast cancer panel. The tumor suppressor genes TP53 as well as BRCA1 and BRCA2 represented the highest mutation rates (>5%) among all BC samples. Interestingly, additional genes potentially playing a pivotal role in BC biology like EP300 were also found to be mutated at a high rate in TNBC. The biological significance of the EP300 gene remains unknown. Additionally, comparing the mutation rates of several genes like TP53, PIK3CA, BRCA2, ATM, RET and EGFR between established BC subtypes showed significant differences.

Conclusion. Next generation sequencing of samples obtained from primary breast cancer tumors confirmed a high level of pathogenic or likely pathogenic mutations in a subtype-specific pattern involving genes like TP53, BRCA1/2, ATM, EGFR, RB1 and PIK3CA.

Citation Format: Bemanian V, Sauer T, Joel T, Katja V, Vessela K, Ida B, Jürgen G. Next generation sequencing (NGS) reveals high mutation rates in all established breast cancer subtypes with subtype-specific patterns [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P1-06-10.