Classification of laryngeal lesions using advanced endoscopic imaging and real-time evaluation by AI - CLAIRE-AI

• Applicants: Professor Dr. Christian Betz; Professor Dr. Gereon Hüttmann; Professor Dr.-Ing. Alexander Schlaefer
• Subject Area: Otolaryngology, Phoniatrics and Audiology ; Methods in Artificial Intelligence and Machine Learning
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 569352564

Tumors of the upper aerodigestive tract (UADT) are the sixth most common tumor entity in humans, with laryngeal cancer accounting for 30-40%. Thereby, an earlier diagnosis goes along with a better prognosis for the patients concerned. As a visual endoscopic assessment is currently not able to replace histopathological examinations, current standard of care in newly detected laryngeal lesions comprises a so-called microlaryngoscopy under general anesthesia combined with invasive tissue sampling, which is costly as well as time consuming and prone to sampling errors. An optical method enabling an on-site distinction between healthy tissue, dysplastic tissue and cancerous tissue in an outpatient setting would therefore have the potential to revolutionize early laryngeal cancer detection. Various optical techniques providing different information on the tissues investigated have been tested for classification of mucosal lesions of the larynx in (pre-) clinical studies and shown promising results, but they do have their individual limitations as standalone procedures. Therefore, multimodal imaging using a combination of techniques has been suggested and trialed, but – mostly owing to its complexity in data acquisition and interpretation – has not yet progressed into clinically meaningful applications. In a new and promising attempt, we aim to combine optical coherence tomography (OCT) and narrow band imaging (NBI) using a flexible endoscope and thus combining depth and spatial information of the suspicious tissue. To overcome the current limitations of multimodal image interpretation, we aim to study different artificial intelligence (AI) methods to fuse information and assist with image analysis following a spatially and temporally correct registration of the multimodal imaging data. Our first goal in this project is the development of a single outpatient endoscopic procedure that allows for an onsite tissue classification in the larynx (“optical biopsy”). As a second goal, the performance of this automated multimodal endoscopic imaging will be evaluated against the current standard-of-care. To achieve these goals, we will design and manufacture biocompatible OCT fiber probes. The probes will fit into the working channel of a flexible NBI-ready nasopharyngolaryngoscope. To evaluate the multimodal imaging, we will perform a study on 60 awake patients with flat mucosal lesions of the endolarynx and 60 healthy volunteers. State-of-the-art deep learning methods will be applied for an automated tissue classification based on both individual imaging modalities and the multimodal data set. This will make our system for computer-assisted diagnosis (CAD) independent of the experience of the physician, reducing the problem of interrater variability.

• DFG Programme: Research Grants