Background: Over half a million individuals are diagnosed with head and neck cancer each year globally. Radiotherapy is an important curative treatment for this disease, but it requires manual time to delineate radiosensitive organs at risk. This planning process can delay treatment while also introducing interoperator variability, resulting in downstream radiation dose differences. Although auto-segmentation algorithms offer a potentially time-saving solution, the challenges in defining, quantifying, and achieving expert performance remain. Objective: Adopting a deep learning approach, we aim to demonstrate a 3D U-Net architecture that achieves expert-level performance in delineating 21 distinct head and neck organs at risk commonly segmented in clinical practice. Methods: The model was trained on a data set of 663 deidentified computed tomography scans acquired in routine clinical practice and with both segmentations taken from clinical practice and segmentations created by experienced radiographers as part of this research, all in accordance with consensus organ at risk definitions. Results: We demonstrated the model’s clinical applicability by assessing its performance on a test set of 21 computed tomography scans from clinical practice, each with 21 organs at risk segmented by 2 independent experts. We also introduced surface Dice similarity coefficient, a new metric for the comparison of organ delineation, to quantify the deviation between organ at risk surface contours rather than volumes, better reflecting the clinical task of correcting errors in automated organ segmentations. The model’s generalizability was then demonstrated on 2 distinct open-source data sets, reflecting different centers and countries to model training. Conclusions: Deep learning is an effective and clinically applicable technique for the segmentation of the head and neck anatomy for radiotherapy. With appropriate validation studies and regulatory approvals, this system could improve the efficiency, consistency, and safety of radiotherapy pathways.

5555 Background: Olfactory neuroblastoma is rare with 25 new cases annually in the UK. Cranio-facial resection and post operative radiotherapy is standard treatment. The use of concomitant chemotherapy has not been fully evaluated, although chemo-sensitivity has been demonstrated by retrospective series. This study evaluates the use of radical radiotherapy and concomitant chemotherapy. Methods: Since 1978 more than 70 patients have been seen at the Institute of Otology and Laryngology. Eighteen patients treated with radiotherapy between 1999 and 2005 were reviewed. These were all CT planned and treated with conformal radiotherapy ± cisplatin chemotherapy 100 mg/m² D 1, 22 and 43. The radiotherapy prescribed dose varied from 50Gy to 62 Gy in 1.8 to 2 Gy fractions, 10 (56%) received 60Gy in 30 fractions. Overall survival, disease free survival and treatment related morbidity were assessed. Results: Seventeen patients were post operative and 1 was neo-adjuvant. Mean age was 46 years (range 20–83) and the majority were Kadish stage C. Of the 17 post-operative patients, 13 (76%) had surgery as primary treatment and 4 (24%) for recurrence. Eight patients (44%) received concomitant chemotherapy. At the time of analysis follow up ranged from 4 months to 72 months (median 17.5, mean 28.2), 16 (88.9%) patients were alive and 14 (77.8%) disease free. Four (22.2%) patients developed recurrence and 2 died of disease. Of the 4 patients with recurrence, 3 did not receive concurrent cisplatin chemotherapy and the fourth patient did not undergo radical surgery. No patients reported deterioration in vision or significant morbidity following treatment. Conclusions: Post operative radiotherapy with concurrent cisplatin chemotherapy is effective and well tolerated in the management of olfactory neuroblastoma. Where concurrent chemotherapy was not given, relapse rates were high. Our results, although immature, suggest disease free survival superior to published data. A prospective multi-centre study to evaluate concomitant treatment schedules is required. No significant financial relationships to disclose.

Background: We evaluated the utility of HNSCC LN R2* relaxation times to infer the oxygenation status of LN non-invasively at baseline and when breathing air and 100% oxygen to predict chemoradiotherapeutic locoregional response at 2 years. Hypoxia within LNs has been associated with poorer outcomes following CRT. Deoxyhaemoglobin decreases MRI transverse relaxation time (T2*) (lengthening inverse, R2*). Methods: A total of 54 patients underwent 1.5T-MRI before CRT. Conventional MR sequences were supplemented with T2* sequences breathing both air and 100% oxygen; pathological nodes identified in consensus were volumetrically contoured to T2* parametric maps. Results: Patients followed-up with for >2 years were categorised by multidisciplinary consensus into post-therapy complete local response (CR; n = 32/54) and local nodal disease relapse (RD; n = 22/54). Our data demonstrated, by R2*, that nodes that sustained post-therapy CR are significantly more hypoxic compared with relapsing nodes and paradoxically demonstrate a significant increase in hypoxia on 100% oxygen. Pre-treatment LN short axis diameter, various qualitative descriptors of malignancy, and quantitative DWI were not useful in discriminating successful response to CRT. Conclusions: This study demonstrates that a significant differential response to 100% oxygen and higher baseline R2* LN measurements could be exploited in risk stratification prior to CRT, and future work could be directed towards understanding the contrast mechanisms of R2* imaging, underpinning the observed differences in the context of hypoxia.

Background: Sinonasal undifferentiated carcinoma (SNUC) is an extremely rare, high-grade, and aggressive tumor of the sinonasal tract. Due to the rarity of this malignancy, current treatment guidelines are based on small and often/mainly single-center retrospective datasets. In the absence of a universally accepted standard of care for SNUC, treatment approaches vary across countries and institutions, reflecting real-world clinical practice. The primary aim of this study was to describe real-world treatment and outcomes for patients with confirmed SNUC. Methods: This was an international, multi-center, retrospective, observational cohort study that pooled patients into the largest SNUC dataset to date. Fifteen centers were enrolled to contribute data, including seven from Europe, four from the United States, three from the United Kingdom, and one from Canada. In the absence of a universally accepted standard of care for SNUC, treatment approaches varied across countries and institutions, reflecting real-world clinical practice. Patients included were those with histologically confirmed SNUC who were treated between 1997 and 2021. Results: This study yielded 485 patients treated for SNUC. The median age at diagnosis was 55.6 years (IQR: 44.5–67.6), and 63.7% were male. Most cases presented at advanced stages, with 70.8% as T4a or T4b. Overall survival (OS) outcomes were available for 412 patients, with a median follow-up of 26.0 months. The 5- and 10-year OS were 47.2% (95% CI: 40.8–53.3%) and 39.6% (95% CI: 32.5–46.6%), respectively. Advanced age, dichotomized T-stage (T4a/b vs. T1–3), M-stage, and orbital involvement were significant poor prognostic factors on univariable analysis (p’s < 0.01). On multivariable analysis, orbital involvement (HR: 2.73, 95% CI: 1.42–5.27, p = 0.003) and distance metastasis stage (HR: 3.00, 95% CI: 1.25–7.21, p = 0.014) were both independently associated with worse OS. Conclusions: This observational study presents the largest multi-center cohort analysis of SNUC to date, providing new insights into prognostic factors for a rare cancer treated at global centers of excellence. Orbital involvement and the presence of metastases are candidate independent risk factors associated with poorer OS.