A Hybrid Neural-Evolutionary Approach for Diagnosing Human Coronary Conditions from their Raman Spectra

P.P.B. de Oliveira, O. Vogler, and C.E. da Matta (Brazil)


Evolutionary computation, neural network, Raman spectra, coronary, atheroma, automatic diagnosis.


Early diagnostic of bad conditions in patients with cardiovascular diseases can dramatically increase their chance of survival. Such an early detection may rely on a laser system that is introduced into the patient's coronary so as to excite its inner walls, and an optical catheter that carries the resulting radiation to a Raman spectrometer at its other end; with the spectra obtained it is then possible to automatically diagnose the condition of the coronary. Here, an approach is described that performs such an automatic diagnostic, based on a genetic algorithm coupled to an artificial neural network. While the neural network is the actual diagnosing system – able to discriminate between three conditions of human coronaries: normal, atheromatous and calcified – the evolutionary algorithm is used to select the spectra frequencies that the neural network should account for. The best networks obtained have achieved optimal performance, a remarkable result that rivals its forerunners from the literature, while preserving a simple solution scheme.

Important Links:

Go Back