Medical background information

Regulation Thermogram

Regulation Thermography is a method of medical diagnosis that utilizes changes in the thermoregulation ability of the skin as an indicator for ongoing diseases.
The skin possesses the ability to control the body's heat loss depending on the ambient temperature through various mechanisms like sweating or adapting the diameter of the capillaries. These mechanisms are governed by specific regions of the brain, which are connected with the skin via nerves. In the spinal cord such nerves can interact with nerve fibres connecting e.g. inner organs with the brain (so-called reflex arc). Thus a disease of an inner organ can influence the thermoregulation ability of a specific area of the skin. Since the human body is innervated in a segmented manner an association of a specific skin region to a certain inner organ is possible.

Reflex arc and Segmented innervation

Measurement

The thermoregulation ability is quantified by measuring the skin temperature at 110 defined points (also called areas) before and after exposing the body to a cold stimulus triggered through undressing. The temperature data thus obtained are plotted in a medically meaningful sequence into a bar chart called >>regulation thermogram<<. The temperatures before and after the cold stimulus of each of the areas are displayed beside each other (black and red bar's in the picture). The measurement itself is performed using a contact thermometer capable of storing the data

 

 

 

Measuring device produced by gbo

Evaluation

Regulation thermograms are evaluated by experts with the help of rules that were extracted from long lasting experience in this field. These rules assess deviations of the observed temperature pattern from the >>normal<< one as well as the reaction to the cold stimulus. Typically in this task groups of areas rather than single ones are investigated. Since the >>normal<< thermogram can only be determined up to natural fluctuations, the expert rules are usually fuzzy in their formulation.
A thermogram comprises 220 values; therefore its evaluation is tedious for the expert and awkward for the newcomer. A computer-based diagnosis aiding system may help in this situation.

Within the scope of the project a diagnosis aiding system based on Fuzzy Logic and Neural Nets is implemented. The system is supposed to be able to classify thermograms on a discrete scale with respect to the presence of patterns specific for female breast cancer. To this end a sufficiently extensive set of expert rules for the evaluation of thermograms is translated into Fuzzy Logic statements, and subsequently is implemented into a so-called Fuzzy Inference System. The process of translation is performed in collaboration with the >>Deutsche Gesellschaft für Onkologie<<. The Fuzzy Inference System allows to apply the Fuzzy calculus to the implemented rules given a concrete thermogram, and yields a description of the presence and intensity of pathological patterns specific for breast cancer.

Beside the rule-based Fuzzy approach Neural Nets are utilized to classify thermograms with respect to pathological patterns. The training of such nets is carried out using thermograms classified by experts. However, a thermogram consisting of 220 values, using the original thermograms as inputs for a Neural Net would lead to an excessive number of neurons and consequently to the necessity to provide large training sets. This is avoided by deploying a subset of the fuzzy expert rules in a preprocessing step that reduces the dimension. A weighted mean of the evaluation results of the Fuzzy Inference System and the Neural Net is taken as the final output of the system. The goodness of the automatic diagnosis aiding thus obtained can be estimated applying methods from nonparametric statistics.

Thermograms classified by experts can also be used to automatically create new evaluation rules. Within the scope of the project Classification Trees, Support-Vector-Maschines and data-driven processes to create fuzzy expert rules are investigated.

Publications

• H. Knaf, Regulation Thermography - a case study of data mining in the medical science, in: A Roadmap for Mathematics in European Industry, MACSI-net, Kaiserslautern 2004.
• H. Knaf, P. Lang, D. Prätzel-Wolters, Expertensysteme in der komplementären Onkologie , in: J.Beuth (Hrsg.), Grundlagen der Komplementäronkologie, Hippokrates, Stuttgart 2002.

• Case Study [ PDF 46 KB ]