In this paper, we build and compare multiple speech systems for the automatic evaluation of the severity of a speech impairment due to oral cancer, based on spontaneous speech. To be able to build and evaluate such systems, we collected a new spontaneous oral cancer speech corpus from YouTube consisting of 124 utterances rated by 100 non-expert listeners and one trained speech-language pathologist, which we made publicly available. We evaluated the systems in two scenarios: a scenario where transcriptions were available (reference-based) and a scenario where transcriptions might not be available (reference-free). The results of extensive experiments showed that (1) when transcriptions were available, the highest correlation with the human severity ratings was obtained using an automatic speech recognition (ASR) retrained with oral cancer speech. (2) When transcriptions were not available, the best results were achieved by a LASSO model using modulation spectrum features. (3) We found that naive listeners’ ratings are highly similar to the speech pathologist's ratings for speech severity evaluation. (4) The use of binary labels led to lower correlations of the automatic methods with the human ratings than using severity scores.

@en

In this paper, we introduce a new corpus of oral cancer speech and present our study on the automatic recognition and analysis of oral cancer speech. A two-hour English oral cancer speech dataset is collected from YouTube. Formulated as a low-resource oral cancer ASR task, we investigate three acoustic modelling approaches that previously have worked well with low-resource scenarios using two different architectures; a hybrid architecture and a transformer-based end-to-end (E2E) model: (1) a retraining approach; (2) a speaker adaptation approach; and (3) a disentangled representation learning approach (only using the hybrid architecture). The approaches achieve a (1) 4.7% (hybrid) and 7.5% (E2E); (2) 7.7%; and (3) 2.0% absolute word error rate reduction, respectively, compared to a baseline system which is not trained on oral cancer speech. A detailed analysis of the speech recognition results shows that (1) plosives and certain vowels are the most difficult sounds to recognise in oral cancer speech — this problem is successfully alleviated by our proposed approaches; (3) however these sounds are also relatively poorly recognised in the case of healthy speech with the exception of/p/. (2) recognition performance of certain phonemes is strongly data-dependent; (4) In terms of the manner of articulation, E2E performs better with the exception of vowels — however, vowels have a large contribution to overall performance. As for the place of articulation, vowels, labiodentals, dentals and glottals are better captured by hybrid models, E2E is better on bilabial, alveolar, postalveolar, palatal and velar information. (5) Finally, our analysis provides some guidelines for selecting words that can be used as voice commands for ASR systems for oral cancer speakers.

@en

Oral cancer speech is a disease which impacts more than half a million people worldwide every year. Analysis of oral cancer speech has so far focused on read speech. In this paper, we 1) present and 2) analyse a three-hour long spontaneous oral cancer speech dataset collected from YouTube. 3) We set baselines for an oral cancer speech detection task on this dataset. The analysis of these explainable machine learning baselines shows that sibilants and stop consonants are the most important indicators for spontaneous oral cancer speech detection.@en

About half of advanced stage head and neck squamous cell carcinoma (HNSCC) patients can be cured by chemoradiotherapy. Patient outcome may be partially determined by the genetic alterations in HNSCC, rendering these alterations promising candidate prognostic factors and/or therapeutic targets. However, their relevance in patient outcome prognosis remains to be assessed in patients that receive standard-of-care chemoradiotherapy. We therefore tested whether frequent genetic alterations were associated with progression free survival (PFS) in advanced stage HNSCC patients who were uniformly treated with definitive platinum-based chemoradiotherapy. To this end, we performed targeted DNA sequencing on frozen pre-treatment tumor biopsy material from 77 patients with advanced stage oro- and hypopharyngeal carcinoma. This provided somatic point mutation and copy number aberration data of 556 genes. The most frequently mutated genes, TP53 (62%), CCND1 (51%), CDKN2A (30%) and PIK3CA (21%), were not associated with PFS. However, co-occurring CCND1 and CDKN2A mutations were associated with short PFS (HR 2.24, p = 0.028) in HPV-negative tumors. Furthermore, tumor mutational burden (sum of somatic point mutations) showed a trend towards decreased PFS (HR 1.9, p = 0.089), and chromosomal instability (CIN) was associated with shorter PFS (HR 2.3, p = 0.023), independent of HPV status. Our results show that tumor mutational burden, CIN markers, and co-occurring CCND1 and CDKN2A mutations are associated with chemoradiotherapy outcomes in advanced stage oro- and hypopharyngeal HNSCC patients, thereby highlighting their prognostic potential. Given their poor prognosis association and link to biological targets, they may also identify patients for novel targeted therapies and immunotherapies

@en

Background: Careful identification of nerves during head and neck surgery is essential to prevent nerve damage. Currently, nerves are identified based on anatomy and appearance, optionally combined with electromyography (EMG). In challenging cases, nerve damage is reported in up to 50%. Recently, optical techniques, like diffuse reflectance spectroscopy (DRS) and fluorescence spectroscopy (FS) show potential to improve nerve identification. Methods: 212 intra-operative DRS/FS measurements were performed. Small nerve branches (1-3 mm), on near-nerve adipose tissue, muscle and subcutaneous fat were measured during 11 surgical procedures. Tissue identification was based on quantified concentrations of optical absorbers and scattering parameters. Results: Clinically comprehensive parameters showed significant differences (<0.05) between the tissues. Classification using k-Nearest Neighbor resulted in 100% sensitivity and a specificity of 83% (accuracy 91%), for the identification of nerve against surrounding tissues. Conclusions: DRS/FS is a potentially useful intraoperative tool for identification of nerves from adjacent tissues. Level of Evidence: Observational proof of principle study.

@en

Background Large cancer genome studies continue to reveal new players in treatment response and tumorigenesis. The discrimination of functional alterations from the abundance of passenger genetic alterations still poses challenges and determines DNA sequence variant selection procedures. Here we evaluate variant selection strategies that select homozygous variants and rare SNPs and assess its value in detecting tumor cells with DNA repair defects. Methods To this end we employed a panel of 29 patient-derived head and neck squamous cell carcinoma (HNSCC) cell lines, of which a subset harbors DNA repair defects. Mitomycin C (MMC) sensitivity was used as functional endpoint of DNA crosslink repair deficiency. 556 genes including the Fanconi anemia (FA) and homologous recombination (HR) genes, whose products strongly determine MMC response, were capture-sequenced. Results We show a strong association between MMC sensitivity, thus loss of DNA repair function, and the presence of homozygous and rare SNPs in the relevant FA/HR genes. Excluding such selection criteria impedes the discrimination of crosslink repair status by mutation analysis. Applied to all KEGG pathways, we find that the association with MMC sensitivity is strongest in the KEGG FA pathway, therefore also demonstrating the value of such selection strategies for exploratory analyses. Variant analyses in 56 clinical samples demonstrate that homozygous variants occur more frequently in tumor suppressor genes than oncogenes further supporting the role of a homozygosity criterion to improve gene function association or tumor suppressor gene identification studies. Conclusion Together our data show that the detection of relevant genes or of repair pathway defected tumor cells can be improved by the consideration of allele zygosity and SNP allele frequencies.@en

Objective: Locally advanced oral squamous cell carcinoma (OSCC) shows lower locoregional control and disease specific survival rates than laryngeal and pharyngeal squamous cell carcinoma (L/P-SCC) after definitive chemoradiotherapy treatment. Despite clinical factors, this can point towards a different tumor biology that could impact chemoradiotherapy response rates. This prompted us to compare the mutational profiles of OSCC with L/P-SCC. Methods: We performed target capture DNA sequencing on 111 HPV-negative HNSCC samples (NKI dataset), 55 oral and 56 laryngeal/pharyngeal, and identified somatic point mutations and copy number aberrations. We next expanded our analysis with 276 OSCC and 134 L/P-SCC sample data from The Cancer Genome Atlas (TCGA dataset). We focused our analyses on genes that are frequently mutated in HNSCC. Results: The mutational profiles of OSCC and L/P-SCC showed many similarities. However, OSCC was significantly enriched for CASP8 (NKI: 15% vs 0%; TCGA: 17% vs 2%) and HRAS (TCGA: 10% vs 1%) mutations. LAMA2 (TCGA: 5% vs 19%) and NSD1 (TCGA: 7% vs 25%) mutations were enriched in L/P-SCC. Overall, we find that OSCC had fewer somatic point mutations and copy number aberrations than L/P-SCC. Interestingly, L/P-SCC scored higher in mutational and genomic scar signatures associated with homologous recombination DNA repair defects. Conclusion: Despite showing a similar mutational profile, our comparative genomic analysis revealed distinctive features in OSCC and L/P-SCC. Some of these genes and cellular processes are likely to affect the cellular response to radiation or cisplatin. Genomic characterizations may guide or enable personalized treatment in the future.

@en