Abstract
At present there is an increase in the elderly population and more persons with cognitive troubles, where dementia is a socio-sanitary challenge. Mild cognitive impairment (MCI) may be a prodromal state of Alzheimer’s disease and other dementias. It is considered an optimal target for diagnosis, likely to be highly prevalent in the future, worldwide. MCI diagnosis is principally based in cognitive and daily living functional activities assessment. However, in clinical settings, essentially primary care setting, MCI is challenging because of time, consulting restrictions and even difficulties understanding cognitive test cut-off points, mainly when diagnosis depends on two or more scales, and it is underdiagnosed. An intelligent system to assist in MCI diagnosis, based on hybrid neural architectures, the counter-propagation network (CPN), with a wrapper approach, has been designed. The dataset includes scores of three commonly used scales, MMSE, GDS and FAQ, along with years of education and age, relative to 203 normal control subjects and 128 subjects who revealed a MCI, from ADNI database. The efficiency of the proposed CPN-based system, with MMSE, FAQ and age, was evaluated using several performance measurements and the clinical utility index (CUI). Its diagnostic performance was compared with a geriatrician, a neurologist and two family physicians. Our proposal achieved the highest score amongst all, AUC: 95,11%, Accuracy: 86,84%, Sensitivity: 90%, Specificity:84,78%, CUI: 0,715. These results were also better than optimum cut-off over each one of the tests. Neural computing methods may be useful tools in clinical settings even when employing brief screening tests.