Prediction of clinical response to excimer laser treatment in vitiligo by using neural network models
Simone Cazzaniga1, Fabrizia Sassi, Santo Raffaele Mercuri
1Centro Studi GISED, Fondazione per la Ricerca Ospedale Maggiore (FROM), presidio ospedaliero Matteo Rota, Bergamo, Italy. cazzaniga_81@hotmail.it
Background:
A predictive model may help to select likely responders and to anticipate treatment duration in vitiligo.
Methods:
We aimed to develop a predictive rule based on data from a randomized trial of excimer laser in vitiligo. Information on 325 treated patches was available. The degree of repigmentation was assessed by digital image analysis of UVB-reflected photographs. Since no strong relationship between any single predictive parameter and outcome was initially documented, we relied on artificial neural networks.
Results:
Using a time-response optimal threshold model, data were divided into 2 groups of responders and nonresponders. A discriminant network was trained in order to detect responders versus nonresponders. A regression network was subsequently used to compute repigmentation time in responders. The neural network discriminator achieved 66.46 +/- 5.37% (95% CI) overall accuracy. The mean absolute error of the neural network regressor was 19.5843 +/- 2.0930 with a root mean square error of 23.7156 +/- 2.2225.
Conclusion:
Our study offers insight into the difficulty of clinical prediction in vitiligo and presents a way to develop an instrument with which to predict the clinical time response in patients treated by excimer laser.