Special computer algorithms can help doctors to make rapid diagnoses of known diseases

Just like urine and blood, breath contains traces of the products of metabolism. Such products can also be signs of infection, inflammation or cancer. For their analysis, computational bioinformatics researchers at the Cluster of Excellence "Multimodal Computing and Interaction" at Saarland University developed special computer algorithms that can help doctors to make diagnoses quickly and reliably. The researchers will be giving a practical demonstration at Booth 34 in Pavilion 26 at Cebit. The computer fair takes place in Hanover from March 6 to 10.

Jan Baumbach is head of the research group "Computational Systems Biology" at the Cluster of Excellence in Saarbrücken. The group studies how to search efficiently and reliably through huge amounts of biomedical data, created through new analysis techniques, with the help of calculation methods from the field of computer science. In cooperation with the Korea Institute for Science and Technology Europe (KIST Europe), the bioinformatics researchers in Saarbrücken analyze doctors' examination results from various medical facilities, including clinics in Hemer, Homburg, Essen, Göttingen and Marburg, among others. Through clinical studies, the doctors analyze the respiration of patients with known diseases, such as lung cancer and infections.

"The technique of measurement has been perfected for several years," Jan Baumbach explains. "Now it's up to computer science to evaluate the measured results." His research group believes in calculation methods which are usually applied for machine learning in the field of artificial intelligence. Using these, they try to find models within the measured products of metabolism, so-called metabolites, which can identify the disease in a body like a fingerprint at a crime scene. "The huge problem is that we have a crime scene with millions of possible indicators, of which maybe only two or three are relevant" says Jan Baumbach. Thus, the bioinformatics researchers leave the decision as to which combination of metabolites indicates a disease to the specially-developed classification algorithms. Using the samples - which, for the human viewer, would be impossible to analyze - the algorithms learn training material that aids in automatically placing unknown data reliably into the categories "healthy" or "disease X."

"Chronic obstructive pulmonary disease (COPD), for example, can be analyzed very accurately, with a failure rate of under 5 percent," says Jan Baumbach. To be able to put the results to practical use, some more clinical studies have to be performed. However, the scientist is convinced of the success of the idea. In five years, he thinks the necessary hardware will fit into a smart phone, replacing the current 18-kilogram machines. With the appropriate algorithms, bacteria and tumors (for example) could be detected more quickly and reliably, blood glucose levels could be measured through breathing into the smart phone.