Head & throat atlas (KOpf-HALs-Atlas) for cancer therapy

KOHALA is a project for developing an innovative system for automated segmentation of organs and structures in the head and throat area, to be used in radiation therapy. In contrast to existing approaches, a new procedure shall be developed and tested, which significantly raises the automatic's success ratio (on more than 90% of all layers) and thus largely avoids manual post-processing of the individual dataset which is common practice today. A manual segmentation of the organs in the head and throat area takes around 80 minutes (60-120 min).

To this end, we work with the new concept of the „articulated atlas“ which has been developed and demonstrated at the Fraunhofer Institute for Computer Graphics Research. The target group for our system are the radiation therapy departments of hospitals, of which approx. 300 exist in Germany alone. For the West European region, which is part of our key market, these numbers may be multiplied by factor ~10.

The milestones consist of data acquisition for „learning“ the articulated atlas of a so-called statistic reference patient, as well as development of the software which finally transfers the generated atlas to an individual patient data set. Clinical validation and the integration of the prototype into the existing "ProSoma " software system represent the last milestone.

Technological realization and stages of development

  1. A head/throat scan of a patient is being uploaded into the system as a DICOM volume.
  2. The scan is first rigidly registered with a so-called articulated head & throat atlas. The articulated head & throat atlas is an innovative model of the head and throat area first published by the IGD (Fraunhofer Institute for Computer Graphics Research), in which single anatomic structures have been recorded statiscally together with their anatomic variability.
  3. Subsequently, a non-rigid registration of the articulated atlas with the patient's scan is carried out. In the process, the atlas' structures are being adjusted and remodeled in an anatomically correct way, so that they correlate with the corresponding structures of the patient's scan. Compared to conventional atlas procedures, only a single atlas is needed in this case, as the anatomic variance is implicitly modeled in the atlas itself.
  4. Subsequently, the marking and designation of the segmented structures in the individual patient scan is achieved, as is common with atlases, via the known position & characteristics of the identical structure within the atlas.
  5. The markings are then excerpted as a DICOM RT Structure and can be uploaded into the radiation treatment planning system to be used for planning.

Functionality of the automatic atlas-based registration system (only bone structures are displayed here for illustration)

Project partners:
MedCom GmbH, Rundeturmstraße 12, 64283 Darmstadt (software development, integration)
Fraunhofer IGD, Fraunhoferstraße 1, 64283 Darmstadt (software development)
Klinikum Offenbach GmbH, Starkenburgring 66, 63069, Offenbach am Main (data acquisition, validation)
Universitätsklinikum Giessen und Marburg, location Marburg (data acquisition, validation)

The project is being supported in the context of the Hessian LOEWE-Förderlinie 3. „Landes-Offensive zur Entwicklung Wissenschaftlich-ökonomischer Exzellenz“, brief: LOEWE, is the title of the research funding program of the German federal state Hessen, which has been existing since 2008.