Migraine is a chronic disease that affects the daily development of
activities of people around the world. To alleviate the symptoms,
OnabotulinumtoxinA (BoNT-A) has solid proven evidence for their use
according to various works and clinical trials. Nowadays, it is known
that 70-80% of patients with chronic migraine show an improvement with
this treatment (improvement defined as a reduction in migraine attack
frequency or days with attacks by at least 50% within 3 months,
leading to a significantly improved functioning of the patients and
their overall quality of life). As has been mentioned by [1], it is
very important to predict if the BoNT-A treatment will be effective in
a patient. Knowing the phenotype-response relationship may help in the
development of new treatments for the 20-30% of patients that do not
respond to the treatment.
This talk will describe two approaches for addressing the prediction
of the therapeutic response to BoNT-A: panoramic and feedback
prediction [2].
Panoramic prediction makes it possible to decide whether the treatment
will be beneficial without using previous knowledge and without
involving unnecessary treatments. Feedback prediction can be more
accurate prediction since it considers the results of previous stages
of the treatment. With the purpose of unveiling the medical attributes
that make treatment effective for patients, consensus models are
applied to the prediction models found through the proposed
approaches. The following attributes have been found to be relevant
when predicting the treatment response to BoNT-A:
migraine time evolution, unilateral pain, analgesic abuse, headache
days and the retroocular component. According to doctors, these
factors are also medically relevant and in alignment with the medical
When training the prediction models, an attribute weighting task is
considered. It is performed with the purpose of finding those weights
that improve the representation of the numeric labels encoded by
doctors for each stage of BoNT-A treatment. In the panoramic
prediction, the attribute weighting is multiobjective because we need
to find the optimal weights that improve the prediction accuracy for
all stages, simultaneously. In this sense, multiobjective evolutionary
algorithms (MOEAs) that support parallelization have been considered
for improving the training time of predictive models [3].
The obtained results show accuracies close to 85% and 90% for
panoramic and feedback prediction approaches, respectively. Moreover,
the training time of the panoramic prediction models is decreased from
8 to less than 2 hours when using 8 threads.


FRANKLIN PARRALES BRAVO received the M.Sc. degree in computer science from the Complutense University of Madrid (UCM), Spain, in 2015, where he received a scholarship to develop the master’s Thesis with the Department of Computer Architecture and Automation (DACYA-UCM) by the Ecuadorian Ministry of Education, Science, Technology and Innovation (SENESCYT) under the 165-ARG5-2013 grant, and the M.Sc. degree in artificial intelligence from the Technical University of Madrid (UPM), Spain, in 2019. He has been granted a predoctoral fellowship by SENESCYT-Ecuador under the 8905- AR5G-2016 grant, to develop his Ph.D. thesis at UCM. Since 2016, he has been an Associate Professor of computer science with the University of Guayaquil, Ecuador, where he is currently focused on data processing methodologies in e-Health for categorizing therapeutic responses in patients with migraine. He is currently with the Department of Computer Architecture and System Engineering, Complutense University of Madrid, Madrid, Spain. His current research interests include e-Health and machine learning.


Date: 2019-Nov-08     Time: 11:00:00     Room: 336

For more information: