Artificial neural network posturography detects the transition of vestibular neuritis to phobic postural vertigo

Abstract

Artificial neural networks (ANNW), described in detail by Duda et al. [1], can be efficiently used to master complex data sets by applying computational analysis for routine clinical uses, for example, to classify the risk of falls in the elderly on the basis of an analysis of balance control during gait [2]. In an earlier study we used ANNW posturography to identify typical postural sway patterns that allow the diagnosis of various balance disorders [3]. Body sway was measured by means of posturography during ten test conditions of increasing difficulty. These included standing with eyes open or eyes closed, with head extended backward, standing on a slab of foam rubber, and tandem stance. Sixteen values were selected from the calculated parameters of each single condition, such as sway path, root mean square values, and Fourier analysis. This means that a total of 160 values were entered into the artificial neural network (for methods, see [3]). In this way a standard three-layer, feed-forward ANNW that uses a backpropagation algorithm was trained with training cases, validated with validation cases, and its accuracy was tested on new cases with various diagnoses. The sensitivity and specificity were about 0.9 for patients with vestibular neuritis or for patients with phobic postural vertigo. Once designed and tested, ANNW-posturography can be considered a black box, which each examiner can apply to predict a specific diagnosis even without a preceding clinical examination. Here we would like to present an example of two patients to demonstrate the clinical relevance of this method. In case 1 the method was able to disclose the transition that a patient with acute vestibular neuritis underwent to phobic postural vertigo within months after disease onset. In case 2 it confirmed the patient’s recovery from static postural control despite the permanent unilateral vestibular loss. Case 1: a 70-year-old male patient presented with typical signs and symptoms of an acute, left-sided vestibular failure. A unilateral deficit of vestibular function was diagnosed by a pathological head-impulse test that revealed the presence of corrective (catch-up) saccades and by caloric irrigation that showed non-responsiveness of the left horizontal semicircular canal. Since there were no other neurological deficits, he was diagnosed to have leftsided vestibular neuritis. This diagnosis was further supported by ANNW posturography with a probability of 93% (Fig. 1a). Dizziness and imbalance recovered gradually within weeks to months due to central compensation, although there was no restitution of peripheral vestibular function. The patient was again able to ski. Five months later, however, he presented with new complaints of subjective postural and gait unsteadiness, which were invisible to the observer. He had attack-like exacerbations of the fear of falling without any real falls. The patient now exhibited T. Brandt (&) Institute for Clinical Neurosciences and Integrated Research and Treatment Center for Vertigo, Balance and Ocular Motor Disorders (IFB), Ludwig Maximilian University, Marchioninistr. 15, 81377 Munich, Germany e-mail: thomas.brandt@med.uni-muenchen.de