doi:10.1016/j.neunet.2005.08.015 
Copyright © 2005 Elsevier Ltd All rights reserved.
A composite neural network model for perseveration and distractibility in the Wisconsin card sorting test
Gülay B. Kaplana, 
, 
, Neslihan S. Şengörb, Hakan Gürvitc, İbrahim Gença and Cüneyt Güzelişd
aInformation Technologies Research Institute, TÜBİTAK-Marmara Research Center, Gebze, 41470 Kocaeli, Turkey
bFaculty of Electrical Electronics Engineering, İstanbul Technical University, Maslak, 80626 İstanbul, Turkey
cBehavioral Neurology and Movement Disorders Unit, Department of Neurology, İstanbul Faculty of Medicine, İstanbul University, Çapa, 34490 İstanbul, Turkey
dDepartment of Electrical and Electronics Eng., Engineering Faculty, Dokuz Eylül University, Buca, 35160 İzmir, Turkey
Received 14 December 2003;
accepted 11 August 2005.
Available online 15 December 2005.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
A composite artificial neural network model is proposed to simulate the performance of the Wisconsin Card Sorting Test. The Wisconsin Card Sorting Test is a test of executive functions where prefrontal deficits are matched to some quantitative measures such as percentage of perseverative errors and number of failures to maintain set. In this work, the proposed model is used to simulate the performances of healthy subjects and patients with prefrontal involvement particularly on these measures. The model is designed in such a way that one of the subsystems, namely, the Hopfield network, serves as the working memory and the other, the Hamming block, as the hypothesis generator. The results show that the proposed relatively simple model is capable of simulating the wide range of the performances of both normal subjects and prefrontal patients on the Wisconsin Card Sorting Test. While lowering the Hamming distance in the Hamming block gave rise to progressively more perseverative responses, changing the threshold vector of the Hopfield network resulted in more set maintenance failures. The former manipulation disrupts the abstraction or mental flexibility and the latter sustained attention or perseverance both of which are the major functions of the prefrontal system.
Keywords: Computational modeling; Prefrontal cortex; Executive functions; Wisconsin card sorting test; Perseveration; Distractibility; Hopfield network; Hamming network
Abbreviations: aCG, anterior cingulate; ANN, artificial neural network; dl-PF, dorsolateral prefrontal; FMS, failure to maintain set; LLS, learning to learn score; OF, orbitofrontal; PFC, prefrontal cortex; RC, response card vector; v, feature vector; T, threshold vector; TC, template card vector; W, weight matrix; WCST, wisconsin card sorting test
Fig. 1. Material used in the WCST (Dehaene, & Changeux, 1991).
Fig. 2. (a) When the exp.'s response is ‘correct’ (b) When the exp.'s response is ‘false’ The modes of the proposed composite neural network model for WCST.
Fig. 3. (a) When the exp.'s response is ‘correct’ (b) When the exp.'s response is ‘false’ The block diagrams corresponding to the activated part of the model.
Fig. 4. Perseverative error percentages with different Hamming distance values and under different Hopfield network thresholds. PE%, perseverative error percentage; n, normal subjects (n=150); and ffp, all focal frontal patients (n=73) in Heaton et al.'s (1993) normative sample. Columns and bars represent the means and standard deviations (SD's) respectively. NM: normative mean (11.8 PE%). The graying rows represent the number of SD's from the mean of the normals (+1 SD=18.9 PE%,+2 SD=26 PE%,+3 SD=33.1 PE%,+4 SD=40.2 PE%,+5 SD=47.3 PE%).
Fig. 5. Failure to maintain set scores under different Hopfield network thresholds and with different Hamming distance values. FMS, failure to maintain set score; n, normal subjects (n=150); and ffp, all focal frontal patients (n=73) in Heaton et al.'s (1993) normative sample. Columns and bars represent the means and standard deviations (SD's), respectively. NM: normative mean (0.8 FMS). The graying rows represent the number of SD's from the mean of the normals (+1 SD=2.1 FMS, +2 SD=3.4 FMS, +3 SD=4.7 FMS).
Table 1.
Coded features
Table 2.
The meaning of the rule vectors
Table 3.
The simulation results of WCST
C, Condition; HD, Hamming distance; HT, Hopfield threshold; CR, number of correct responses; CC, Completed categories; PE%, perseverative error percentage; FMS, failure to maintain set score; numbers after±are standard deviations (SD's).
Abstract
Purchase PDF (276 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (1)
Purchase PDF (197 K)
