Aims: This study aims to assess the validity and educational benefit of a 3D printed pituitary training model in the education of otolaryngology and neurosurgery trainees in pituitary adenoma resection, carotid injury and CSF leak management. Methodology: A 3D printed model of the paranasal sinuses, pituitary fossa and internal carotid arteries was used in an educational course for 12 otolaryngology and neurosurgery registrars and consultants. Trainees were guided through 4 standardised educational scenarios - soft adenoma resection, fibrous adenoma resection, carotid injury and CSF leak repair. Three experienced skull base neurosurgeons and 3 experienced skull base otolaryngologists graded the anatomical fidelity, tissue realism and learning benefit of the models on a 5-point Likert scale. During dissection, participants’ heart rate was monitored to assess the model’s ability to simulate the high-stress environment in vascular injury. Trainees completed pre- and post-course questionnaires assessing their degree of comfort with skull base surgery and subjective post-course improvement. Results: All expert neurosurgeons and ENT surgeons agreed or strongly agreed that the model’s anatomical features, tissue realism, and learning goals were beneficial to trainees. Participants’ heart rate showed a significant rise from baseline during the carotid injury (91.8+/-22.5 vs 118.4+/-22.3, p=0.024). There was also a significant decrease in heart rate between the first and second carotid injury simulation (121.0+/-22.5 vs 108.5+/-23.4, p=0.006). All participants reported a self-rated ‘better’ or ‘much better’ improvement in skills in pituitary surgery. There was a significant improvement in self-rated ability to manage vascular injury pre- vs post-course (2.0+/-0.6 vs 3.2+/-0.5, p<0.001). Conclusions: A 3D printed model of the pituitary fossa provides a high-fidelity training environment for simulation of adenoma resection and management of complications.

Aims: Artificial intelligence (AI) algorithms can automate the process of detecting pathology and anatomical variants on radiographic images. This study seeks to determine if a trained Convolutional Neural Network (CNN) AI algorithm can accurately identify lateral lamella of the cribriform plate (LLCP) which are taller than 8 millimetres. This is usually defined as Keros Stage 3, which, while uncommon (around 2% of the population), increases the risk of skull base breach during endoscopic sinus surgery. Methodology: Coronal slices paranasal sinus CT scans were reviewed by two blinded surgical trainees and the height of the LLCP measured (ground truth). These images and measurements were used to train a multi-layer CNN algorithm via supervised learning. The algorithm’s ability to measure the LLCP height was tested using an independent set of images and relevant performance measures were calculated. Results: A total of 500 images were used for training the CNN algorithm and a further 150 images were used for testing. The CNN algorithm measured 81.4% of LLCPs accurately to within 1mm of the ground truth with an overall mean absolute error of 0.61 mm (95% CI: 0.52 - 0.70mm) and mean absolute percentage error of 15.8%. When classified according to the Keros staging system, accuracy was 73.6% (95% CI: 72.9 – 74.5) and area under the curve was 0.79 (95% CI: 0.71 – 0.86). Notably, the algorithm correctly classified TBC% of Keros Stage 3 images correctly. Conclusion: This study demonstrates that CNN AI algorithms can measure clinically important structures, such as the LLCP, on preoperative sinus CT scans. These measurements can be used to classify anatomical variants according to established staging systems to aid in operative planning. Importantly, these algorithms can be refined to ensure rare, high-risk anatomy is appropriately identified for further attention from clinicians.