Amsterdam – the Netherlands – First published study of a deep learning model for the real-time endoscopic detection and classification of benign and malignant lesions across the glottic, supraglottic and hypopharyngeal regions, conducted with Radboudumc and UMC Groningen.
WSK Medical is pleased to announce the publication of a new multicenter study in the peer-reviewed journal Cancers evaluating Zeno AI™, the first CE-MDR approved AI-solution on the market for the real-time detection and classification of head and neck cancers and benign lesions. The study, “Real-Time-Capable Detection of Glottic, Supraglottic and Hypopharyngeal Lesions Using Artificial Intelligence During Flexible Endoscopy”, was conducted by the departments of Otorhinolaryngology and Head and Neck Surgery of Radboud University Medical Center (Radboudumc) in Nijmegen and University Medical Center Groningen (UMCG), with additional cases contributed by the Martini Hospital in Groningen. The Zeno AI™ algorithm was developed in collaboration with WSK Medical.
Supraglottic and hypopharyngeal carcinomas are among the most aggressive head and neck cancers and are often diagnosed at an advanced stage, when treatment is more invasive and the prognosis is poor. Early lesions in these anatomically complex regions are easily missed during routine flexible endoscopy, even by experienced clinicians. In this study, the previously validated Zeno AI™ model for glottic (vocal cord) lesions was expanded into a unified model covering the glottic, supraglottic and hypopharyngeal regions.
The model was trained and tested on a database of 1,336 lesion videos and 123 healthy control videos collected between 2012 and 2024, comprising more than 142,000 video frames, annotated on the basis of histopathologically or clinically confirmed diagnoses. On an independent test set, Zeno AI™ achieved a lesion detection precision of 92.1% and a recall of 73.3%, and at least one correct lesion detection was present in 98.2% of lesion videos. Among correctly detected lesions, malignancy was identified with a sensitivity of 94.3%. Running at 76 frames per second, the model is fully real-time capable during flexible endoscopy. According to the authors, this is the first study to report a deep learning model for the real-time endoscopic detection and classification of benign and malignant laryngeal and pharyngeal lesions.
Example of Zeno AI detecting a hypopharyngeal malignancy during a routine endoscopy.
“The publication of this study in Cancers marks an important milestone for Zeno AI™,” said Marius Wellenstein, CEO of WSK Medical. “By expanding real-time detection and classification beyond the vocal cords to the supraglottic larynx and hypopharynx – regions where cancer is often found at an advanced stage – we are taking the next step in our mission to assist clinicians in detecting and classifying head and neck cancer at earlier stages, when treatment is most effective.”
The authors note that the detection of early-stage (T1) tumors remains an important limitation and a primary focus for further development, as early detection is a key aim of AI-assisted endoscopy. Further multicenter data collection, external validation and prospective studies comparing AI-assisted with conventional endoscopy are the next steps towards clinical deployment. The research was funded by the KWF Dutch Cancer Society and Health Holland.
The full open-access publication is available in Cancers: https://doi.org/10.3390/cancers18162609. If you would like to learn more about Zeno AI™, please contact us!
About WSK Medical
WSK Medical is a global leading Artificial Intelligence software solution provider that focuses on the development and deployment of AI software solutions for the early detection and classification of cancer. Our aim is to bring you the most advanced AI technologies to assist clinicians and doctors. We do this by working in partnership, to develop high performing and secure AI software as a medical device for oncology purposes. For more information about our AI solutions, please contact us.
