Research projects to support medicine and the health sector in the field of diagnosis
DVAT (Dynamic Visual Acuity Testing)
A prototype was developed for the quantitatively reproducible testing of the vestibular reflex. This brainstem reflex causes a compensatory movement of the eyes during rapid head rotations. Vestibular disturbances (dizziness symptoms), for example as a result of a craniocerebral trauma or an inflammatory disease of the vestibular organ, can be measured as reduced visual acuity, which is determined with a moving head, but otherwise classically (for example, by means of a visual sign projector).
For the reliability of the test result it is crucial that the visual sign to be recognized is only presented in a predetermined range of rotation speed. Conventional examination methods are based on a rotation of the patient’s head as instructed by the examiner and are therefore only reproducible to a limited extent. The newly developed prototype measures the rotational speed of the patient’s head and presents visual signs (randomly selected Landolt rings) only when the predetermined rotational speed range is reached. By systematically varying the size of the visual signs presented, the influence of the head rotation on visual acuity is reproducibly determined.
University Hospital Regensburg, Department of Audiology of the ENT Clinic
Device Software & Signal Processing, Prof. Dr. Axel Doering
Robotic Scene Segmentation
Robotic Scene Segmentation” uses artificial intelligence to detect medically relevant objects in endoscopic videos. In the case of a surgical robot such objects can be e.g. the instruments of a robot (manipulator, joint, shaft), anatomical objects (kidney, intestine, …) or medical material (needles, sutures, clamps, …). The recognition and differentiation of the objects is done by a so-called “semantic segmentation”, i.e. each place in the image is assigned an object category. Thus, after the application of automated analysis, it is known where which objects are located in the image.
The semantic segmentation is based on a deep neural network (deep learning) in the form of an encoder-decoder architecture. In this type of network, the encoder is used to extract distinctive features from the input image that are essential for the task. The resulting coded representation of the image is then converted into a mask image by the decoder. This mask image specifies the corresponding object category for each position in the input image.
ReMIC, Prof. Dr. Christoph Palm
Barrett`s Esophagus – Deep Learning for computer-assisted early detection of esophageal cancer from endoscopic images
Reflux is an inflammation of the esophagus, which is caused by an increased reflux of acid stomach contents into the esophagus. Chronic reflux is the main cause of the Barrett’s esophagus, a lesion of the mucous membrane with increased risk of developing esophageal cancer. The survival chances of affected patients are considered poor, as the disease is usually diagnosed at a late stage. If a standard drug treatment of reflux is not successful, an endoscopic examination may be indicated to detect treatable symptoms as early as possible. However, this is not unproblematic, since many reflux patients are endoscopically negative, i.e. mucous membrane lesions are not visible despite the presence of disease (low sensitivity of the examination). The significance in case of a pathological finding is relatively high (high specificity of the examination).
In diagnostic imaging procedures, machine learning methods are increasingly being used. With the help of deep learning approaches, the physician should be supported in reliably recognizing reflux-related mucous membrane damage, especially (pre-)carcinogenic lesions, when evaluating endoscopic images. Based on the machine evaluation of endoscopic images, conclusions on the severity of a possible disease should be drawn. Through the use of Deep Learning, a quality in the diagnostic evaluation of medical images has been achieved several times in recent years that not only reaches the medical “gold standard”, but even exceeds it. This means that physician and computer meet at eye level, so that in the future, for example, the computer could at least be established as a second assessor.
Laboratory for Technology Assessment and Applied Ethics (OTH Regensburg)
University Hospital Augsburg, III. Medical Clinic
São Paulo State University, Brazil
ReMIC, Prof. Dr. Christoph Palm