Kenneth Gilhuijs

UMC Utrecht

In short, Kenneth Gilhuijs's research entails the following: 

Triage of Patients with High-risk Oral Squamous cell Carcinoma using Artificial Intelligence (TOSCA)

Oral squamous cell carcinoma (OSCC) is a type of cancer that affects the mouth and throat. It is relatively rare, accounting for 1-4% of all cancers in western countries with 971 new cases in the Netherlands in 2021. The presence of lymph node metastases - or the spread of cancer to the lymph nodes - is the most important factor in predicting the outcome of the disease.

Currently, patients at risk of occult - or hidden - lymph node metastases are treated with an elective neck dissection (END), a surgical procedure that removes lymph nodes from the neck. However, this procedure turns out to be unnecessary in 80% of early OSCC patients, exposing these patients to unnecessary morbidity.

An alternative is a sentinel lymph node biopsy (SNB), which involves removing and testing a single lymph node for cancer. If the sentinel node is negative, no further treatment is needed. If the sentinel node is positive, a second, more extensive, surgery is necessary (completed neck dissection), requiring repeat hospital admission and general anesthesia. Moreover, this second surgery is more difficult to perform after an elective neck dissection, risking more morbidity and delaying follow-up (adjuvant) treatment.

In this retrospective observational cohort study with approximately 1.000 patients from seven hospitals, we aim to develop a new strategy based on deep learning of MRI to refer the largest number of patients with occult lymph node metastases directly to completed neck dissection, instead of sentinel lymph node biopsy, without referring node-negative patients to neck dissection.

TOSCA is a collaboration between the Image Sciences Institute and the departments Head-and-neck surgical oncology, Radiology, and Radiotherapy of UMC Utrecht. In addition, six other head-and-neck oncology treatment centers in the Netherlands participate.

Brief summary of progress / results July 2024

In the first year of the TOSCA study, two databases were constructed following nWMO study approval by the METC of UMC Utrecht. The databases contain information from consecutively included pseudonymized patients between 2012 and 2021 with early oral squamous cell carcinoma (OSCC) who underwent sentinel node procedures (SNP). Inclusion criteria were: first primary (i.e., no previous OSCC), and early stage (i.e., cT1N0Mx and cT2N0Mx). Exclusion criterion was absence of pre-operative MRI.

The first database is a DICOM imaging database containing preoperative MRI scans. Each MRI examination contains multiparametric series including proton-density STIR, T2, contrast-enhanced T1, T1 SPIR, and diffusion-weighted imaging (DWI). The second database is a CASTOR database with clinicopathological parameters such as patient age, tumor localization, histopathological parameters and outcome. We have currently included 148 patients from UMC Utrecht with 41 occult lymph node metastases (28%). The paper work with the other participating hospitals is in different stages of completion, ranging between the signing of data transfer agreements (DTAs) and actual data transfer.

Methodologically, the first part of this project focused on automated segmentation of the index cancer from multiparametric MRI using deep learning (U-Net) – for which manual delineations were performed in a training cohort – and multiple-instance learning.