Please cite this article in press as: Groznik V, et al. Elicitation of neurological knowledge with argument-based machine learning. Artif Intell Med (2012), http://dx.doi.org/10.1016/j.artmed.2012.08.003 ARTICLE IN PRESS G Model ARTMED-1244; No. of Pages 12 Artificial Intelligence in Medicine xxx (2012) xxx–xxx Contents lists available at SciVerse ScienceDirect Artificial Intelligence in Medicine jou rn al h om epage: www.elsevier.com/locate/aiim Elicitation of neurological knowledge with argument-based machine learning Vida Groznik a,∗ , Matej Guid a , Aleksander Sadikov a , Martin Moˇ zina a , Dejan Georgiev b , Veronika Kragelj c , Samo Ribariˇ c c , Zvezdan Pirtoˇ sek b , Ivan Bratko a a Artificial Intelligence Laboratory, Faculty of Computer and Information Science, University of Ljubljana, Trˇ zaˇ ska cesta 25, SI-1000 Ljubljana, Slovenia b Department of Neurology, University Medical Centre Ljubljana, Zaloˇ ska cesta 2, SI-1000 Ljubljana, Slovenia c Faculty of Medicine, University of Ljubljana, Vrazov trg 2, SI-1104 Ljubljana, Slovenia a r t i c l e i n f o Article history: Received 11 July 2012 Received in revised form 7 August 2012 Accepted 19 August 2012 Keywords: Argument-based machine learning Knowledge elicitation Decision support systems Parkinsonian tremor Essential tremor a b s t r a c t Objective: The paper describes the use of expert’s knowledge in practice and the efficiency of a recently developed technique called argument-based machine learning (ABML) in the knowledge elicitation pro- cess. We are developing a neurological decision support system to help the neurologists differentiate between three types of tremors: Parkinsonian, essential, and mixed tremor (comorbidity). The system is intended to act as a second opinion for the neurologists, and most importantly to help them reduce the number of patients in the “gray area” that require a very costly further examination (DaTSCAN). We strive to elicit comprehensible and medically meaningful knowledge in such a way that it does not come at the cost of diagnostic accuracy. Materials and methods: To alleviate the difficult problem of knowledge elicitation from data and domain experts, we used ABML. ABML guides the expert to explain critical special cases which cannot be handled automatically by machine learning. This very efficiently reduces the expert’s workload, and combines expert’s knowledge with learning data. 122 patients were enrolled into the study. Results: The classification accuracy of the final model was 91%. Equally important, the initial and the final models were also evaluated for their comprehensibility by the neurologists. All 13 rules of the final model were deemed as appropriate to be able to support its decisions with good explanations. Conclusion: The paper demonstrates ABML’s advantage in combining machine learning and expert knowl- edge. The accuracy of the system is very high with respect to the current state-of-the-art in clinical practice, and the system’s knowledge base is assessed to be very consistent from a medical point of view. This opens up the possibility to use the system also as a teaching tool. © 2012 Elsevier B.V. All rights reserved. 1. Introduction and motivation Essential tremor (ET) is one of the most prevalent movement disorders [1]. It is characterized by postural and kinetic tremor with a frequency between 6 and 12 Hz. Although it is regarded as a symmetrical tremor, ET usually starts in one upper limb and then spreads to the other side affecting the contralateral upper limb, consequently spreading to the neck and vocal cords, giving rise to the characteristic clinical picture of the disorder. However, there are many deviations from this classical presentation of ET, e.g. bilateral tremor onset, limb tremor only, head tremor only, ∗ Corresponding author. Present address: Faculty of Computer and Information Science, University of Ljubljana, Trˇ zaˇ ska cesta 25, SI-1000 Ljubljana, Slovenia. Tel.: +386 1 4768987; fax: +386 1 4768386. E-mail address: vida.groznik@fri.uni-lj.si (V. Groznik). URL: http://www.ailab.si/vida (V. Groznik). isolated voice tremor. Parkinsonian tremor (PT), on the other hand, is a resting tremor classically described as “pill rolling” tremor with a frequency between 4 and 6 Hz. It is one of the major signs of Parkinson’s disease (PD), which also includes bradykinesia, rigidity and postural instability. PT is typically asymmetrical, being more pronounced on the side more affected from the disease onset. Although distinct clinical entities, ET is very often misdiagnosed as PT [2]. Results from clinical studies show that ET is correctly diag- nosed in 50–63%, whereas PT in 76% of the cases. Co-existence of both disorders is also possible [3]. In addition, PT can be very often observed when the upper limbs are stretched (postural tremor) and even during limb movement (kinetic tremor), which further complicates the differential diagnosis of the tremors. Digitalized spirography is a quantitative method of tremor assessment [4], based on spiral drawing by the patient on a digi- tal tablet. In addition to precise measurement of tremor frequency, spirography describes tremors with additional parameters – these, together with physical neurological examination, offer new means 0933-3657/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.artmed.2012.08.003