CC BY-NC-ND 4.0 · Laryngo-Rhino-Otol 2018; 97(S 02): S84-S85
DOI: 10.1055/s-0038-1640005
Poster
Onkologie: Oncology
Georg Thieme Verlag KG Stuttgart · New York

Development of an aut. image analysis-method by deep-learning-methods for the detection of head and neck cancer based on stand. real-time near-infrared ICG fluorescence endoscopy-images (NIR-ICG-FE)

A Dittberner
1  Klinik für Hals-, Nasen- und Ohrenheilkunde FSU Jena, Jena
,
S Sickert
2  Lehrstuhl für Digitale Bildverarbeitung FSU Jena, Jena
,
J Denzler
2  Lehrstuhl für Digitale Bildverarbeitung FSU Jena, Jena
,
O Guntinas-Lichius
1  Klinik für Hals-, Nasen- und Ohrenheilkunde FSU Jena, Jena
,
T Bitter
1  Klinik für Hals-, Nasen- und Ohrenheilkunde FSU Jena, Jena
,
S Koscielny
1  Klinik für Hals-, Nasen- und Ohrenheilkunde FSU Jena, Jena
› Author Affiliations
In Kooperation mit dem Lehrstuhl für digitale Bildverarbeitung der FSU Jena, mit Unterstützung von KARL STORZ, Tuttlingen.
Further Information

Publication History

Publication Date:
18 April 2018 (online)

  

Introduction:

Improving the gold standard in the diagnosis of head and neck cancer using white light and invasive biopsy with digital image recognition procedures, there is a need for a development of new technologies. In the sense of an "optical biopsy", they in vivo and online should provide additional objective information for decision making for the head and neck surgeon. Artificial neural networks in combination with machine learning might be a helpful and fast approach.

Material and Methods:

NIR ICG-FE was standardized in patients with head and neck cancer. Video documented standardized tissue biopsies from the carcinomas and their adjacent mucosa were performed. Only histologically proven images were then manually annotated pixel-accurate for the automatic analysis. Training and testing of the automated image recognition algorithm by deep learning methods has been done in the Leave-One-Patient-Out-procedures. As an architecture for the neural network the widespread "AlexNet-Konfiguration" has been used.

Results:

The "AlexNet-Konfiguration" as already trained neural network was suitable to initialize our deep learning method. The newly designed neural network has been successfully refined step by step for carcinoma detection. Previous results allow the conclusion that an automatic detection of head and neck cancer is possible with the method used.

Conclusions:

Image data by NIR ICG-FE can be used in an artificial neural network by deep learning methods to create an automated image recognition algorithm for the detection of head and neck cancer.