Diagnosis And Prognosis Of Urological Diseases using Neural Networks

The deseases diagnosis and prognosis are usually realized by analyses and processing of clinical information. When the volume and the variety of the information become too demanding for the clinician, the need for supportive statistical prediction methods emerges. When the classical methods, like statistical modeling, are failing, due to the computational complexity and to the long processing time, the neural networks (NN) could offer effective solutions, being able to perform real-time prediction of the deseases diagnosis and prognosis of a particular patient. We have developed and validated a neural integrated set of programs, in an adequate programming medium, capable to offer solutions to some urological problems. The inputs are variables carefully selected, obtained from the real situations and readily comparable with the real, functional, clinical models. Models of clinical urological applications have been developed using various NN architectures, such as multilayers perceptrons, radial basis function NN, competitive NN and recurrent NN. A comparison of the performance of different NN architectures and training algorithms has been accomplished and the model with the best accuracy/complexity ratio was selected, in each particular case. For instance, the neural network performance prediction in prostate cancer prediction with a global percentage of correct classification of 96.94 % (calculated as the proportion of correctly classified patients) was better with 2.05 % than one obtained with the statistical logistic regression. This in an encouraging result. In clinical terms this is beneficial, avoiding over treatment in the cases with prostate capsule penetration, when radical prostatectomy is not an option. This also has a beneficial psychological impact on the patient, by avoiding unnecessary surgery. The performance limits of the neural network prediction, in our opinion, was given by the rather reduced dimension of the database and the modality of its collecting.

Selected Publications

• Botoca Corina, R.Bardan, M.Botoca, F.Alexa Prediction of Prostate Capsule Penetration using Neural Networks Recent Advances in COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS and CYBERNETICS (CIMMACS '09) Puerto De La Cruz, Canary Islands, Spain,December 14-16, 2009, 28-33 ISSN: 1790-5117, ISBN: 978-960-474-0352 http://www.wseas.us/conferences/2009/tenerife/eed/
• Botoca Corina, R.Bardan, M.Botoca, F.Alexa Organ Confinement of Prostate Cancer. Neural Networks Assisted Prediction Meditech Cluj, 23-26 september 2009, IFMBE Proceedings, Springer Verlag Series287-290, ISSN: 1680-0737, http://www.meditech.utcluj.ro, http://www.springer.com/series/7403?detailsPage=titles