Abstract

In the clinical routine of hospitals, vast amounts of image and non-image data are generated and stored in a multitude of hospital information systems. The majority of the data is used once for diagnosis and treatment of a specific patient and maybe, a second time for monitoring the patient. The data then, is archived and rests untouched until the legal retention period ends. This represents a huge neglected data treasure, which, if lifted, can contribute to the screening, early diagnosis and therapy of diseases. In contrast to new prospective studies, it can already convey the temporal course of diseases that develop over years or decades. On the other hand, the clinical data were not acquired having comparability across patients in mind, which poses many challenges on data harmonization, cleansing, pre-processing, and analysis.

In this talk, I will present efforts at the Department of Neurology, Otto-von-Guericke-University Magdeburg, Germany targeted at a continuous registration and analysis of brain structures and functions of all patients with neurological and psychiatric diseases in the federal state of Saxony-Anhalt. This will allow the creation of a globally unique database that is of great interest to international biomedical research and industry and can be analyzed with big data technologies to gain further insight into diagnoses and therapies. The prerequisite for this is a consistent digitization of brain data from clinical routine at the highest level. This includes both, a lifting of the data acquired in the past and the ongoing digitization of new data. It requires the establishment of a processing pipeline between data acquisition and evaluation based on components to be developed for harmonization of data, analysis and improvement of data quality, automated extraction of quantitative data characteristics as well as for ensuring patient data protection. In the context of (image) biomarker discovery in dementia research, I will demonstrate the current state of the pipeline and how machine learning techniques support the pipeline steps.

Organizational Details

A Machine Learning-Based Pipeline for Processing Clinical Cohort Study Data

Date: Thursday, 19th December 2019
Time: 10:00 am
Room: O27/331