KOHALA: digital student for cancer treatment

Shortening a time-consuming procedure from four hours to five minutes and automate it at the same time sounds like a dream come true for employees in all fields and industry sectors. This dream could soon become a reality for radiologists. Software could take away the tedious processing of CT images, which is required before cancer radiation therapy.


Photo: User interface; Copyright: Fraunhofer IGD

The user-interface of KOHALA. Different organs are displayed in color: the spinal cord (dark blue), a saliva gland (green) and the larynx (light blue); ©Fraunhofer IGD

The reason for all the preparation is the complexity of specific body regions. Many important organs and structures are particularly located in the head and neck area: large blood vessels, the larynx, glands and lymph nodes are in very close proximity next to the spine, brain stem and spinal cord. This does not just make surgeries quite difficult, it also complicates radiation therapy, so that tumors are difficult to treat in this area. If radiation is too inaccurate, physicians also damage healthy tissue that surrounds the tumor. This is why radiation therapy in this area needs to be meticulously planned.

A series of CT images that show the body layer by layer forms the basis for a three-dimensional model for radiation: "During the first step of radiation treatment planning, radiologists mark organs and structures in the area that is supposed to be irradiated. Even experienced physicians need approximately four hours per patient to do this," Dr. Stefan Wesarg of the Fraunhofer Institute for Computer Graphics (Fraunhofer IGD) explains the digital segmentation of the head and neck. "Our software automates this process and makes it easier."

The software’s abbreviated name is KOHALA (from the German KOpf-HALs-Atlas, meaning a head-neck atlas for radiation therapy) and is supposed to prepare image data to enable radiation treatment planning for tumors. Presently, KOHALA is being tested at the Sana Clinical Center in Offenbach, Germany, and the University Medical Center Gießen and Marburg. "Early results are promising; the feedback from hospitals is positive," says Wesarg. "We assume that physicians will be able to continue the use of segmented images directly for radiation treatment planning." In this case, not just the tedious process of segmentation would be eliminated, but possibly even post-processing on the computer by a person. Only the tumor still needs to be manually marked before the angle, form and ray intensity can be individually defined and optimized.
Photo: KOHALA; Copyright: Fraunhofer IGD

Three-dimensional model of head and neck from KOHALA; ©Fraunhofer IGD

Doctors train the computer

Behind KOHALA is an algorithm that "knows" where the structures are located in the head and neck area. Wesarg and his team let their algorithm benefit from teachers with practical experience just like a regular medical student would: "Physicians, who manually marked the relevant anatomical structures in the image data, participated in its development. We developed our statistical model by taking advantage of this expertise." Now the model is supposed to be able to tell where the different structures are located in relation to each other and how location and form can deviate from this arithmetic mean in the case of individual patients. Based on general data, it is supposed to be applicable again in the field to the individual anatomy of the individual patient.

Even though tests are already running with real patient data in both participating hospital facilities, KOHALA is not part of the actual treatment yet. Instead, the physicians are now checking up on their "student" before it is allowed to make its own decisions. Clinical tests are going to still take approximately half a year, during which it will become clear how much KOHALA really learned. Just like a real medical student, KOHALA could still have problems in some areas where a person needs to take corrective action, as Wesarg points out: "It could be that we need to improve the algorithm with certain complex organs or manually rework the result. However, based on tests that we have performed so far, we are not able to make a concluding statement yet."
Photo: Work at the KOHALA project; Copyright: Fraunhofer IGD

Dr. Wesarg (in the back) and a colleague operating the KOHALA software;
©Fraunhofer IGD

Within the scope of the LOEWE funding line No. 3 ("Förderlinie 3, The State Campaign for the Development of Scientific and Economic Excellence"), the German state of Hessen is also a part of this project along with the Fraunhofer IGD and the hospital facilities in Offenbach and Gießen-Marburg. First the MEDCOM Company in Darmstadt, which is specialized in imaging and radiation treatment planning and is originally a spin-off of the Fraunhofer IGD, is scheduled to implement the software.

The interest in the student's success is therefore huge, and if the student is successful, a continuation of its career is also possible. Automated radiation treatment planning would also make sense for tumors in the stomach and chest areas. "The complexities of the neck/head and the stomach and chest areas are similar. Even though the structures of chest and stomach are larger, so is the corresponding area," explains Wesarg. He therefore considers an expansion of the KOHALA approach to other body regions a potential project sequel.

Background: imaging and radiation treatment planning in video

Certain tumors require systematic radiation treatment planning and precise irradiation. This is the case for instance when the tumor is difficult to access during surgery, because it is located at a risk structure such as the brain or spine, or if the patient becomes sick again and has already received radiation therapy in the past. At any rate, healthy tissue needs to be exposed to a particularly low dose of radiation.

The MEDICA.de report on the German CyberKnife-Center Soest shows which steps are parts of treating cancer with radiation therapy. The CyberKnife is a robotic system for precise cancer radiation. Contrary to other systems, it dynamically balances patient movements (arbitrary and respiratory movements) during treatment, so that the patient does not need to be strapped down.
Video: Surgical robots: highly precise and gentle

Source: MEDICA-tradefair.com
Photo: Timo Roth; Copyright: B. Frommann

© B. Frommann

The article was written by Timo Roth and translated from German by Elena O'Meara.