Aims: To develop a deep learning-based hierarchical model for otoscopic image classification, designed to augment telehealth workflows. To explore a multi-stage sequential analysis to evaluate otoscopic images for quality, external acoustic canal obstruction, tympanic membrane perforations, otitis media, and hearing loss. Methods: Telehealth data aggregated from two sources (Royal Darwin Hospital, NT and Deadly Ears Program, Qld), consisting of training, validation, and test hold-out sets. Five unique classification models were designed, each employing various pre-trained convolutional neural network (CNN) architectures. The models were developed to address specific aspects of workflows to evaluate otoscopic images based on specific formatting, anatomical, and pathological characteristics. To enhance diagnostic precision, an innovative approach was adopted where all base models were integrated into an overall ensemble model, including age and gender. This ensemble model facilitated a sequential analysis of otoscopic images, allowing each stage of the model to contribute to the final classification. Results: The ensemble-workflow model achieved an overall validation accuracy of 90%. In Stage 1, the model achieved 96% accuracy to screen input images based on quality. In Stage 2, the model achieved 95% accuracy to screen images with or without external acoustic canal obstruction. In Stage 3, the model achieved 92% accuracy to identify tympanic membrane (TM) perforations. In Stage 4, the model achieved 85% accuracy to identify otitis media. In Stage 5, the model achieved 84% accuracy to identify conductive hearing loss. Conclusions: The deep learning-based hierarchical model demonstrated high efficacy in multi-stage otoscopic image classification. This approach has the potential to advance ear health and hearing screening by integrating artificial intelligence (AI) into daily workflows to enhance decision-making and triage.

Introduction: Nuclear protein in testis (NUT) sinonasal carcinoma is a recently characterised, rare, and highly aggressive malignancy affecting the midline of the body. Initially classified as a variant of undifferentiated differentiated SCC or SNUC in H&N region, however recent classifications identify it as a distinct rearrangement involving the NUT gene. Despite therapeutic strategies, including surgery, radiation therapy, and chemotherapy, the prognosis for NUT carcinoma remains grim. Current research seeks to unveil NUT fusion-induced oncogenic mechanisms, opening avenues for targeted therapies for this challenging carcinoma. This abstract presents two cases of sinonasal NUT carcinoma. Case 1: A 39-year-old male presenting with frontal forehead soft tissue swelling growing over 3 months, resulting in decreased GCS and required bifrontal decompression of tumour arising from frontal sinus, confirmed to be NUT tumour. Due to tumour extension and rapid progression, resection was deemed impractical. Radiotherapy was planned but due to swift deterioration, patient was palliated and passed away one-month post-diagnosis. Case 2: A 28-year-old female presented with 9-week history of left sided facial pain, epiphora, epistaxis. She underwent left endoscopic sinus surgery and noted NUT tumour in the left antrum with osseous destruction of maxillary sinus. She received palliative radiotherapy and 3 cycles of palliative chemotherapy. This was complicated by febrile neutropenia, immune induced hepatitis, and covid-19. She developed wide bony metastasis to sacrum, thoracolumbar spine, and medial clavicle. She was palliated following deterioration and passed away 6 months post diagnosis. Conclusion: These cases stress the urgent need for innovative therapies in treating aggressive sinonasal NUT carcinoma, highlighting the disease's complexity. Ongoing collaborative research is imperative to advance understanding and treatment strategies for this rare malignancy.

A healthy microbiome is crucial to preventing the development and recurrence of sinus infections. Genomics technologies, including amplicon sequencing and metagenomics have improved our ability to identify difficult-to-culture bacteria, many of which populate the sinonasal microbiome. However, there is no sinus-appropriate mock-community available to test these methods. These positive controls are essential to defining the accuracy of microbiome data and eliminating bias at all stages in the genomics workflow. We created the first positive control relevant to the sinonasal microbiome to benchmark genomic methods of analysis of sinonasal microbiome samples. Growth curves comparing optic density to colony forming units were generated for the 9 bacteria in the mock community. Using these linear regression models, the mock was assembled with equal proportions of each bacterium. Library preparation was done by three different methods: (1) ONT 16S LR amplicon sequencing, (2) full length 16S amplicon sequencing and (3) shotgun sequencing. All samples were sequenced on the ONT GridION platform and taxonomic profiles were generated with EMU for the 16S datasets, and with Sourmash for the metagenomic datasets. Benchmarking shows that current gold-standard method using 16S amplification introduces excessive bias and thus, subsequent taxonomic profiling misrepresents the microbiome. In the controls tested, 16S amplification resulted in an excess of amplicons from S. Aureus and K. Pneumoniae, and an underrepresentation of amplicons from Corynebacterium spp. By contrast, metagenomic sequencing was able to, repeatedly and accurately, recapitulate the taxonomic profiles of the input mock community standards. When we tested these methods on patient samples, we saw a dramatic difference in the taxonomic profile of the microbiome, with shotgun sequencing revealing the dominance of Corynebacterium spp. This work is the first to prove a validated end-to-end workflow for sinus metagenomics.

Background: The review of surgical mortality is a critical aspect of health care quality control and a mandatory for Fellows of Royal Australasian College of Surgeon (RACS) continuing professional development program. The Australian and New Zealand Audit of Surgical Mortality (ANZASM) keeps an audit of all surgical deaths. This peer-reviewed process scrutinises factors contribuiting to patient death and identifies area of concern. As such, we conducted a review of death following Otolaryngology and Head and Neck (OHN) surgery with the hope of identifying common themes. Methods: Data of all patients who had died following OHN surgery death, which was captured by ANZASM from its inception in January 2010 to February 2023 was reviewed. Peer-reviewed reports were reviewed and where there were areas of concern, a thematic approach was used to analyse areas of concern. Results: Our review identified 908 cases of death following OHN surgery took place and of those 154 patients were highlighted for areas of concern. Amongst these concerns 21 (13.6%) cases had issues pre-operatively, 17 (11.0%) had concerns Intra-operatively, and 29 (18.8%) had concerns post-operatively. Communication issues was identified in 10 (6.4%) cases, and 8 (5.2%) patients had delay in recognising complications. Out of the 74 patients who had an adverse event during their care, 21 (28.3%) cases died as result of the adverse event, while 38 (51.4%) may have died as result of the adverse event. Of these, 8 (10.8%) cases were preventable and 33 (44.6%) were probably preventable. Conclusion: This study provides the first large scale review of surgical mortality in all of OHN surgery in Australia. The insights provided through this review aims to highlight pitfalls in care of OHN patients with the aim of helping provide consideration by treating surgeon and thus improve patient safety.

There is evolving interest in the role that AI technology may play in improving access to ear healthcare service delivery through the detection of middle ear disease using otoscopic images. Studies have shown that medical practitioners and physicians have favourable opinions of the impact of AI on healthcare, however, there have not been any studies exploring the perceptions of potential end-users of AI technology that assists in the diagnosis of middle ear disease. The successful clinical translation of new technology is dependent on end-user uptake and thus understanding barriers may be helpful to encourage end-user engagement with AI technologies. Aim: To explore the perceptions of AI technology to diagnose middle ear disease using otoscopic images amongst clinical audiologists and community hearing screeners. Method: This study involved three focus groups (n=30 participants) conducted with NZ paediatric audiologists and community-based vision hearing screeners. Thematic analysis was used to analyse data. Result: AI technology was perceived to have no impact on the roles of audiologists or hearing screeners but has potential to streamline workflow and reduce workload. Additionally, both audiologists and hearing screeners felt it could build confidence and additional skillsets thereby improving clinical decision making. AI was seen as a useful tool to better engage with patients and encourage adherence with management plans. There was general consensus that AI could improve access to ear healthcare services particularly in rural or remote settings. Barriers to uptake included concerns regarding data privacy, potential for misdiagnosis and clinical risk. Conclusion: AI technology is perceived to benefit potential end-users both from a professional perspective and workload efficiency. AI technology was also seen to provide benefit to healthcare services and engagement with patient groups. Further work is recommended to overcome perceived barriers to uptake of new AI technology.

Aims Emissions in surgery is a relatively unexplored area; and in particular, is yet to be explored in detail in ENT. We wanted to quantify the greenhouse gas (GHG) emission from a tonsillectomy procedure and assess the proportion of each source’s contribution. Method Operational data from tonsillectomies performed at Queen Elizabeth University Hospital (Glasgow) were gathered and converted to global warming potential using established conversion factors and data from existing healthcare-focused carbon footprint studies. The domains considered were waste, pharmaceuticals, surgical instrument decontamination, transportation, consumables and utilities. This study used a process-based carbon footprint approach based on the ‘Greenhouse Gas Protocol: Product Life Cycle Accounting and Reporting Standard. Results The carbon footprint of a typical case was 41 kgCO2e which is equivalent to driving a car for approximately 150 miles (equivalent of a return car journey between Auckland and Hamilton). Consumables were responsible for 17% of this; 14% came from transport, 5.4% from decontamination, 4.8% from pharmaceuticals and 4% from waste. However, the largest GHG was from utilities, of which heating, ventilation and air conditioning was the overwhelming contributor. Conclusion The largest sources of GHG emissions here will require hospital-wide initiatives. However, there are aspects of consumables and waste streams we can still improve on in ENT surgery. These include the move from disposable to reusable instruments as well as increasing availability and use of recycling waste streams in theatres. This study has now helped provide a template that can be applied to other ENT procedures for analysis and will feed into sustainability working groups within the organisation.