Project 2

INNOVATIVE ARTIFICIAL INTELLIGENCE METHODS FOR IMPROVED TUMOR DIAGNOSTICS USING OPTICAL TECHNIQUES

To date, there is no adequate screening procedure for the early detection of squamous cell carcinoma of the upper air-duct or head and neck region (HNSCC), which in their entirety constitute the sixth most common tumor entity in humans. This not infrequently leads to a delay in diagnosis with consecutive worsening of the prognosis with further increased costs in terms of follow-up care and rehabilitation. The newly founded working group "Innovative Optical Diagnostic Procedures" (Head: Dr. D. Eggert) under the new clinic director and chair of the ENT clinic Prof. C. Betz deals with the application and further development of different optical procedures for improved diagnostics in HNSCC. Promising methods for improved early detection of HNSCC are optical coherence tomography (OCT), narrow band imaging (NBI) and hyperspectral imaging (HSI), which are available at the ENT of the UKE. These methods have in common that the evaluation of the obtained image data is complex. Furthermore, the interpretation of the image data strongly depends on the evaluating physician/scientist, which leads to a high interobserver variability, especially in difficult situations. Machine learning algorithms are often applied to evaluate image data in a reproducible manner. In particular, convolutional neural networks (CNN) have become a very promising approach for classification and segmentation of image data. Under the direction of Prof. Dr. A. Schlaefer, the Institute of Medical Engineering Systems at the TUHH is intensively involved in the computer-aided, automated evaluation of such medical image data, in particular also for cancer detection using optical modalities. A particular challenge and current subject of research continues to be the adaptation of the methods to medical image modalities and clinical issues, especially in the context of small amounts of data.

Hypotheses: The following hypotheses will be tested within the PhD:

1. modern optical methods like OCT, NBI, HSI or fluorescence endoscopy are suitable for early detection of HNSCC.

2. artificial intelligence methods (especially CNN) can be used to evaluate image data reproducibly, quickly and with clinically relevant accuracy.

Objectives: The goal of the project is to combine innovative artificial intelligence methods with modern optical techniques to automatically detect and label normal tissue, hyperplasia, dysplasia and HNSCC in the image data. New approaches for the selection and adaptation of suitable learning architectures based on physical measurement principles and clinical data will be developed.