Deep learning algorithm to detect vocal cord cancer live during an endoscopy

A new study published in Head&Neck in August 2023 has found that a self-developed deep learning algorithm is able to localize and classify malignant and benign vocal cord lesions live during an endoscopy. David Wellenstein and Guido van den Broek of the department of Otorhinolaryngology and Head and Neck Surgery of the Radboudumc, together with WSK Medical BV, have created this algorithm in order to assist ENT surgeons (in training) during the diagnostic trajectory in patients with a vocal cord lesion.

Currently, patients with a suspicious vocal cord lesion either undergo follow-up or a diagnostic procedure under general anesthesia to obtain histology. Both have implications for the patient. By aiding the clinician live during an endoscopy, the goals are to increase the detection rate of (pre)malignant vocal cord lesions in an early stage and limit the number of unnecessary diagnostic procedures performed under general anesthesia.

This study has shown that the algorithm is able to correctly localize a vocal cord lesion and is able to correctly classify a malignant vocal cord lesion in 71% tot 78% of the cases. Furthermore, it is able to function live during an endoscopy in the outpatient clinic, meaning that it responds while the ENT surgeon (in training) is performing the endoscopy. This algorithm is created as an extra tool to assist the clinician in the outpatient clinic during the diagnostic trajectory, but cannot replace the clinical judgement of the clinician. Currently, a prospectively multicenter study is initiated to investigate the accuracy and feasibility of the algorithm in the outpatient clinic. Furthermore, other anatomical subsites of the throat, and in a later phase of adjacent areas (e.g. esophagus, oral cavity), will be added to the algorithm.

Read the study here

Wellenstein DJ, Woodburn J, Marres HAM, van den Broek GB. Detection of laryngeal carcinoma during endoscopy using artificial intelligence. Head Neck. 2023 Sep;45(9):2217-2226. doi: 10.1002/hed.27441. Epub 2023 Jun 28. PMID: 37377069.