# A Novel Approach for Cognitive Clustering of Parkinsonisms through Affinity Propagation

^{1}

^{2}

^{3}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Participants

#### 2.2. MMSE Assessment

#### 2.3. Statistical Analysis

#### 2.3.1. Residuals Calculation

**Y**(n × m), the matrix of the MMSE subscale scores, and

**X**(n × k) the matrix of covariates, the multiple linear regression equation is:

**X**(k) (independent variables), ${b}_{j}^{0}$ is the intercept, ${b}_{j}^{1}$, …, ${b}_{j}^{k}$ are the coefficients and ${e}_{ij}$ is the model deviation. Then, we obtained the residual ${e}_{ij}$ of the i-th observation of the j-th MMSE subscale score as:

**E**(n × m) was then used as input for the cluster analysis through affinity propagation, as described in the next section.

#### 2.4. Affinity Propagation

_{ij}], where n is the number of participants to be clustered and i and j are the participant’s indices. A commonly used measure of distance or dissimilarity is the squared Euclidean distance. If there are two points in an m dimensional space where m is the number of measured variables (the eleven MMSE subscales), then the squared Euclidean distance is defined as:

_{iz}is the value of the z-th variable for the i-th residual and e

_{ik}is the value of the z-th variable for the j-th residual in

**E**(n × m), the matrix of regression residuals, calculated as described in Section 2.3.1. Larger values of s

_{ij}indicate a greater degree of similarity between i and j than smaller values of s

_{ij}. It is assumed that s

_{ii}= 0 (for all 1 ≤ i ≤ n) in the formulation. In the present work, the negative squared Euclidean distance was used, as suggested by Frey and Dueck [28].

_{i}, for 1 ≤ i ≤ n, which may also be specified as the n × 1 vector

**p**). In other words, the preferences represent an a priori knowledge of how good a participant could be eligible as an exemplar [29]. Usually, all samples are equally eligible to be an exemplar, thus their value p is the same constant [8,29]. Anyhow, the preferences vector is none other than a control parameter and increasing the value of one sample results in an increase in the likelihood that the sample will be chosen as exemplar [28,29]. The default specification for the preferences is the median of the similarity values (

**S**

^{M}) [8,28], although it has been shown that by using the median, the number of clusters could be dramatically overestimated [8,46]. A reliable alternative is to specify the preference as the minimum of the similarity measures: p

_{i}=

**S**

^{min}, for 1 ≤ i ≤ n, resulting in a smaller number of clusters [28]. For assessing which between the median and the minimum of the similarity matrix is to prefer in the case of the clustering of the MMSE scores, we used both preference values and then we compared their findings with a Silhouette analysis [47] as described in Section 2.5.

**R**) and the availability matrix (

**A**) [8,28,30]. The final results are stored in a fourth matrix named criterion matrix (

**C**). These three matrices are iteratively updated by the following four equations:

**ξ**, containing the labels which are none other than the exemplars (clusters) to which each participant (point) belongs.

Algorithm 1 | Cognitive clustering of regression residuals trough Affinity Propagation |

Input: | MMSE subscale scores matrix Y(n × m), covariates matrix X(k), damping factor d |

Output: | Number of exemplars ex, cluster assignment vector ξ |

1: | initialization: E(n × m), matrix of regression residuals; S(n × n), matrix of similarities; p(n × 1), preference vector of Affinity Propagation; |

2: | forj = 1 to m do |

3: | fit_{j} = ols(y_{j} ~ x_{1} + … + x_{k}); fitted model of the j-th column in Y, with x_{1}, …, x_{k} as the columns in X |

4: | for i = 1 to n−1 do |

5: | e_{ij} = y_{ij}—fit_{j}.predict(y_{ij}); regression residual = the difference between y_{ij} and its prediction by fit_{j} |

6: | fori = 1 to n do |

7: | for j = 1 to m do |

8: | for z = 1 to m do |

9: | s_{ij} = −$\sum _{\mathrm{z}=1}^{\mathrm{m}}{\left({e}_{iz}-{e}_{jz}\right)}^{2}$; negative squared Euclidean distance between e_{iz} and e_{jz} |

10: | p_{i} = S^{min} OR p_{i} = S^{median} |

11: | ap = AffinityPropagation(S, p, d); |

12: | ex = size(ap.exemplars); |

13: | ξ = ap.cluster_membership; |

#### 2.5. Clustering Accuracy Assessment

**p**. For this reason, we compared different clustering results obtained by using the minimum and the median of the similarity matrix. A commonly used approach to assess the quality of clustering results is the Silhouette index, which is a measure of how close each point in a cluster is to the points in its neighboring cluster. Given a cluster X

_{j}(j = 1, …, c where c is the number of clusters found), the Silhouette index Sil

_{i}(i = 1, …, n where n is the number of points in the cluster) of the i-th point in the cluster X

_{j}is given by:

_{j}, mavd(i) is the mean minimum distance between the i-th point and all the points in the cluster X

_{k}(k = 1, …, c), and max is the maximum operator. The Silhouette values range varies from −1 to 1, where a value close to 1 could mean that the i-th point was accurately clustered to the optimum exemplar of the cluster. A value near to 0 could suggest that the i-th point could be attributed to the nearest cluster, whereas a value near to −1 could mean that the i-th point was misclassified [47]. Thus, for characterizing the heterogeneity and isolation of a cluster Y

_{j}(j = 1, …, c) it is possible to calculate the cluster Silhouette S

_{ij}l given by:

_{ij}l. The mean value of all the Silhouette indices of all the points in a partition $U\leftrightarrow Y:{Y}_{l}U\dots {Y}_{i}U\dots {Y}_{c}$, is called Global Silhouette and it is given by the equation:

## 3. Results

#### 3.1. Statistical Analysis

#### 3.2. Cluster Analysis

- Cluster #1 CTRL: male, age 62, education 16, MMSE subscales = [5, 5, 3, 5, 3, 2, 1, 3, 1, 1, 1];

- Cluster #1 PD: male, age 43, education 13, MMSE subscales = [5, 5, 3, 4, 3, 2, 1, 3, 1, 1, 1];
- Cluster #2 PD: female, age 61, education 8, MMSE subscales = [5, 5, 3, 5, 3, 2, 1, 3, 1, 1, 1];

- Cluster #1 PSP: male, age 57, education 17, MMSE subscales = [5, 5, 3, 2, 1, 2, 1, 3, 0, 1, 0];
- Cluster #2 PSP: male, age 78, education 13, MMSE subscales = [5, 5, 3, 5, 2, 2, 1, 3, 0, 1, 0];

- Cluster #1: CTRL, female, age 59, education 16, MMSE subscales = [5, 5, 3, 5, 3, 2, 1, 3, 1, 1, 1];
- Cluster #2: PD, female, age 61, education 8, MMSE subscales = [5, 5, 3, 5, 3, 2, 1, 3, 1, 1, 1];
- Cluster #3: PD, female, age 70, education 5, MMSE subscales = [1, 2, 3, 1, 1, 2, 1, 3, 0, 0, 0];
- Cluster #4: PSP, male, age 78, education 8, MMSE subscales = [4, 4, 3, 1, 1, 2, 1, 3, 0, 1, 0].

## 4. Discussion

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## Appendix A

## Appendix B

**Table A1.**Results of the multiple linear regression with the method ordinary least squares (ols) by accounting for the influence of age, sex and education years on the eleven MMSE subscales of the whole cohort comprising of CTRL, PD and PSP participants. The parameters are the R squared, the F value, the intercept β

^{0}, the coefficients of age, sex and education years, (β

^{1}, β

^{2}, β

^{3}) as well as the p-value.

R^{2} | F | Intercept (β ^{0}, p-Value) | Age (β ^{1}, p-Value) | Sex (β ^{2}, p-Value) | Education (β ^{3}, p-Value) | |
---|---|---|---|---|---|---|

TO | 0.193 | 10.8 | 4.9218, <0.001 | −0.0162, 0.104 | −0.1011, 0.563 | 0.0739, <0.001 |

OS | 0.309 | 20.3 | 4.0724, <0.001 | −0.0142, 0.110 | 0.1743, 0.264 | 0.0983, <0.001 |

Reg | 0.074 | 3.6 | 3.0925, <0.001 | −0.0029, 0.331 | −0.0489, 0.344 | 0.0118, 0.039 |

AC | 0.312 | 20.6 | 3.6018, 0.004 | −0.0267, 0.080 | −0.0646, 0.809 | 0.1675, <0.001 |

Rec | 0.114 | 5.8 | 2.9579, <0.001 | −0.0159, 0.092 | −0.1284, 0.437 | 0.0426, 0.020 |

N | 0.083 | 4.1 | 1.7130, <0.001 | 0.0010, 0.598 | 0.0586, 0.085 | 0.0106, 0.005 |

SR | 0.022 | 1 | 0.7640, 0.004 | 0.0007, 0.821 | −0.0240, 0.674 | 0.0098, 0.119 |

P | 0.071 | 3.4 | 3.1548, <0.001 | −0.0069, 0.089 | 0.0584, 0.410 | 0.0111, 0.153 |

W | 0.252 | 15.3 | 1.2908, <0.001 | −0.0125, 0.003 | −0.0789, 0.278 | 0.0283, <0.001 |

CE | 0.150 | 8 | 0.9553, 0.004 | −0.0054, 0.183 | −0.0788, 0.270 | 0.0253, 0.010 |

D | 0.379 | 27.7 | 1.3470, <0.001 | −0.0180, <0.001 | 0.0741, 0.271 | 0.0321, <0.001 |

## References

- Tryon, R.C. Cluster analysis. Edwards Brothers. Ann. Arbor Mich.
**1939**, 122. [Google Scholar] - Le, N.Q.K.; Do, D.T.; Chiu, F.Y.; Yapp, E.K.Y.; Yeh, H.Y.; Chen, C.Y. XGBoost Improves Classification of MGMT Promoter Methylation Status in IDH1 Wildtype Glioblastoma. J. Pers. Med.
**2020**, 10, 128. [Google Scholar] [CrossRef] - Ho Thanh Lam, L.; Le, N.H.; Van Tuan, L.; Tran Ban, H.; Nguyen Khanh Hung, T.; Nguyen, N.T.K.; Huu Dang, L.; Le, N.Q.K. Machine Learning Model for Identifying Antioxidant Proteins Using Features Calculated from Primary Sequences. Biology
**2020**, 9, 325. [Google Scholar] [CrossRef] - Johnson, S.C. Hierarchical clustering schemes. Psychometrika
**1967**, 32, 241–254. [Google Scholar] [CrossRef] - Ward, J.H., Jr. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc.
**1963**, 58, 236–244. [Google Scholar] [CrossRef] - Rao, M. Cluster analysis and mathematical programming. J. Am. Stat. Assoc.
**1971**, 66, 622–626. [Google Scholar] [CrossRef] - Steinhaus, H. Sur la division des corp materiels en parties. Bull. Acad. Pol. Sci.
**1956**, 1, 801. [Google Scholar] - Brusco, M.J.; Steinley, D.; Stevens, J.; Cradit, J.D. Affinity propagation: An exemplar-based tool for clustering in psychological research. Br. J. Math. Stat. Psychol.
**2019**, 72, 155–182. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Allen, D.N.; Goldstein, G. Cluster Analysis in Neuropsychological Research; Springer: Berlin/Heidelberg, Germany, 2013. [Google Scholar]
- Morris, R.; Blashfield, R.; Satz, P. Neuropsychology and cluster analysis: Potentials and problems. J. Clin. Exp. Neuropsychol.
**1981**, 3, 79–99. [Google Scholar] [CrossRef] - Cotrena, C.; Damiani Branco, L.; Ponsoni, A.; Milman Shansis, F.; Paz Fonseca, R. Neuropsychological Clustering in Bipolar and Major Depressive Disorder. J. Int. Neuropsychol. Soc.
**2017**, 23, 584–593. [Google Scholar] [CrossRef] [PubMed] - Hermens, D.F.; Hodge, M.A.R.; Naismith, S.L.; Kaur, M.; Scott, E.; Hickie, I.B. Neuropsychological clustering highlights cognitive differences in young people presenting with depressive symptoms. J. Int. Neuropsychol. Soc.
**2011**, 17, 267–276. [Google Scholar] [CrossRef] [PubMed] - Burdick, K.; Russo, M.; Frangou, S.; Mahon, K.; Braga, R.; Shanahan, M.; Malhotra, A. Empirical evidence for discrete neurocognitive subgroups in bipolar disorder: Clinical implications. Psychol. Med.
**2014**, 44, 3083. [Google Scholar] [CrossRef] [Green Version] - Lewandowski, K.; Sperry, S.; Cohen, B.; Öngür, D. Cognitive variability in psychotic disorders: A cross-diagnostic cluster analysis. Psychol. Med.
**2014**, 44, 3239. [Google Scholar] [CrossRef] [Green Version] - Kim, S.Y.; Lim, T.S.; Lee, H.Y.; Moon, S.Y. Clustering mild cognitive impairment by mini-mental state examination. Neurol. Sci.
**2014**, 35, 1353–1358. [Google Scholar] [CrossRef] [PubMed] - Phillips, M.L.; Stage, E.C., Jr.; Lane, K.A.; Gao, S.; Risacher, S.L.; Goukasian, N.; Saykin, A.J.; Carrillo, M.C.; Dickerson, B.C.; Rabinovici, G.D.; et al. Neurodegenerative Patterns of Cognitive Clusters of Early-Onset Alzheimer’s Disease Subjects: Evidence for Disease Heterogeneity. Dement. Geriatr. Cogn. Disord.
**2019**, 48, 131–142. [Google Scholar] [CrossRef] [PubMed] - Alashwal, H.; El Halaby, M.; Crouse, J.J.; Abdalla, A.; Moustafa, A.A. The Application of Unsupervised Clustering Methods to Alzheimer’s disease. Front. Comput. Neurosci.
**2019**, 13, 31. [Google Scholar] [CrossRef] [PubMed] - Van Rooden, S.M.; Heiser, W.J.; Kok, J.N.; Verbaan, D.; Van Hilten, J.J.; Marinus, J. The identification of Parkinson’s disease subtypes using cluster analysis: A systematic review. Mov. Disord.
**2010**, 25, 969–978. [Google Scholar] [CrossRef] - Mu, J.; Chaudhuri, K.R.; Bielza, C.; de Pedro-Cuesta, J.; Larrañaga, P.; Martinez-Martin, P. Parkinson’s Disease Subtypes Identified from Cluster Analysis of Motor and Non-motor Symptoms. Front. Aging Neurosci.
**2017**, 9. [Google Scholar] [CrossRef] - Liu, P.; Feng, T.; Wang, Y.-j.; Zhang, X.; Chen, B. Clinical heterogeneity in patients with early-stage Parkinson’s disease: A cluster analysis. J. Zhejiang Univ. Sci. B
**2011**, 12, 694. [Google Scholar] [CrossRef] - van Rooden, S.M.; Colas, F.; Martinez-Martin, P.; Visser, M.; Verbaan, D.; Marinus, J.; Chaudhuri, R.K.; Kok, J.N.; van Hilten, J.J. Clinical subtypes of Parkinson’s disease. Mov. Disord.
**2011**, 26, 51–58. [Google Scholar] [CrossRef] - Erro, R.; Vitale, C.; Amboni, M.; Picillo, M.; Moccia, M.; Longo, K.; Santangelo, G.; De Rosa, A.; Allocca, R.; Giordano, F.; et al. The heterogeneity of early Parkinson’s disease: A cluster analysis on newly diagnosed untreated patients. PLoS ONE
**2013**, 8, e70244. [Google Scholar] [CrossRef] - Ma, L.Y.; Chan, P.; Gu, Z.Q.; Li, F.F.; Feng, T. Heterogeneity among patients with Parkinson’s disease: Cluster analysis and genetic association. J. Neurol. Sci.
**2015**, 351, 41–45. [Google Scholar] [CrossRef] - Pont-Sunyer, C.; Hotter, A.; Gaig, C.; Seppi, K.; Compta, Y.; Katzenschlager, R.; Mas, N.; Hofeneder, D.; Brücke, T.; Bayés, A. The Onset of Nonmotor Symptoms in Parkinson’s disease (The ONSET PD S tudy). Mov. Disord.
**2015**, 30, 229–237. [Google Scholar] [CrossRef] - Jin, X.; Han, J. K-Means Clustering; Springer: Berlin/Heidelberg, Germany, 2010. [Google Scholar]
- Steinley, D.; Brusco, M.J. Evaluating mixture modeling for clustering: Recommendations and cautions. Psychol. Methods
**2011**, 16, 63. [Google Scholar] [CrossRef] [PubMed] - Milligan, G.W.; Cooper, M.C. An examination of procedures for determining the number of clusters in a data set. Psychometrika
**1985**, 50, 159–179. [Google Scholar] [CrossRef] - Frey, B.J.; Dueck, D. Clustering by passing messages between data points. Science
**2007**, 315, 972–976. [Google Scholar] [CrossRef] [Green Version] - Mézard, M. Where are the exemplars? Science
**2007**, 315, 949–951. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Thavikulwat, P. Affinity Propagation: A Clustering Algorithm for Computer-Assisted Business Simulations and Experiential Exercises. Available online: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.490.7628&rep=rep1&type=pdf (accessed on 3 February 2021).
- Hoglinger, G.U.; Respondek, G.; Stamelou, M.; Kurz, C.; Josephs, K.A.; Lang, A.E.; Mollenhauer, B.; Muller, U.; Nilsson, C.; Whitwell, J.L.; et al. Clinical diagnosis of progressive supranuclear palsy: The movement disorder society criteria. Mov. Disord.
**2017**, 32, 853–864. [Google Scholar] [CrossRef] - Quattrone, A.; Morelli, M.; Nigro, S.; Quattrone, A.; Vescio, B.; Arabia, G.; Nicoletti, G.; Nistico, R.; Salsone, M.; Novellino, F.; et al. A new MR imaging index for differentiation of progressive supranuclear palsy-parkinsonism from Parkinson’s disease. Parkinsonism Relat. Disord.
**2018**, 54, 3–8. [Google Scholar] [CrossRef] [PubMed] - Barbagallo, G.; Morelli, M.; Quattrone, A.; Chiriaco, C.; Vaccaro, M.G.; Gulla, D.; Rocca, F.; Caracciolo, M.; Novellino, F.; Sarica, A.; et al. In vivo evidence for decreased scyllo-inositol levels in the supplementary motor area of patients with Progressive Supranuclear Palsy: A proton MR spectroscopy study. Parkinsonism Relat. Disord.
**2019**, 62, 185–191. [Google Scholar] [CrossRef] [PubMed] - Goldman, J.G.; Holden, S.K.; Litvan, I.; McKeith, I.; Stebbins, G.T.; Taylor, J.P. Evolution of diagnostic criteria and assessments for Parkinson’s disease mild cognitive impairment. Mov. Disord.
**2018**, 33, 503–510. [Google Scholar] [CrossRef] [Green Version] - Postuma, R.B.; Berg, D.; Stern, M.; Poewe, W.; Olanow, C.W.; Oertel, W.; Obeso, J.; Marek, K.; Litvan, I.; Lang, A.E.; et al. MDS clinical diagnostic criteria for Parkinson’s disease. Mov. Disord.
**2015**, 30, 1591–1601. [Google Scholar] [CrossRef] - Vaccaro, M.G.; Sarica, A.; Quattrone, A.; Chiriaco, C.; Salsone, M.; Morelli, M.; Quattrone, A. Neuropsychological assessment could distinguish among different clinical phenotypes of progressive supranuclear palsy: A Machine Learning approach. J. Neuropsychol.
**2020**. [Google Scholar] [CrossRef] [PubMed] - Carpinelli Mazzi, M.; Iavarone, A.; Russo, G.; Musella, C.; Milan, G.; D’Anna, F.; Garofalo, E.; Chieffi, S.; Sannino, M.; Illario, M.; et al. Mini-Mental State Examination: New normative values on subjects in Southern Italy. Aging Clin. Exp. Res.
**2020**, 32, 699–702. [Google Scholar] [CrossRef] [PubMed] - Folstein, M.F.; Folstein, S.E.; McHugh, P.R. “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J. Psychiatr. Res.
**1975**, 12, 189–198. [Google Scholar] [CrossRef] - Grigoletto, F.; Zappalà, G.; Anderson, D.W.; Lebowitz, B.D. Norms for the Mini-Mental State Examination in a healthy population. Neurology
**1999**, 53, 315. [Google Scholar] [CrossRef] - Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. Scikit-learn: Machine learning in Python. J. Mach. Learn. Res.
**2011**, 12, 2825–2830. [Google Scholar] - Harris, C.R.; Millman, K.J.; van der Walt, S.J.; Gommers, R.; Virtanen, P.; Cournapeau, D.; Wieser, E.; Taylor, J.; Berg, S.; Smith, N.J. Array programming with NumPy. Nature
**2020**, 585, 357–362. [Google Scholar] [CrossRef] [PubMed] - Hunter, J.D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng.
**2007**, 9, 90–95. [Google Scholar] [CrossRef] - McKinney, W. Data structures for statistical computing in python. In Proceedings of the 9th Python in Science Conference, Austin, TX, USA, 28 June–3 July 2010; pp. 51–56. [Google Scholar]
- Seabold, S.; Perktold, J. Statsmodels: Econometric and statistical modeling with python. In Proceedings of the 9th Python in Science Conference, Austin, TX, USA, 28 June–3 July 2010; p. 61. [Google Scholar]
- Hutcheson, G.D. Ordinary least-squares regression. In L. Moutinho and GD Hutcheson, the SAGE Dictionary of Quantitative Management Research; Sage: Southend Oaks, CA, USA, 2011; pp. 224–228. [Google Scholar]
- Brusco, M.J.; Köhn, H.-F. Exemplar-based clustering via simulated annealing. Psychometrika
**2009**, 74, 457–475. [Google Scholar] [CrossRef] - Rousseeuw, P.J. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J. Comput. Appl. Math.
**1987**, 20, 53–65. [Google Scholar] [CrossRef] [Green Version] - Dueck, D. Affinity Propagation: Clustering Data by Passing Messages. Ph.D. Thesis, University of Toronto, Toronto, ON, Canada, September 2009. [Google Scholar]
- Brusco, M.J.; Shireman, E.; Steinley, D. A comparison of latent class, K-means, and K-median methods for clustering dichotomous data. Psychol. Methods
**2017**, 22, 563–580. [Google Scholar] [CrossRef] [PubMed] - Starczewski, A. A new validity index for crisp clusters. Pattern Anal. Appl.
**2017**, 20, 687–700. [Google Scholar] [CrossRef] [Green Version]

**Figure 1.**Schematic representation of the functioning of the affinity propagation algorithm in a 3D fashion. Figure adapted and revised from Frey and Dueck [28].

**Figure 2.**Radar plots of the mean of the eleven Mini Mental State Examination (MMSE) subscales per diagnosis: healthy controls (CTRL), Parkinson’s disease patients (PD) and Progressive Supranuclear Palsy patients (PSP). In gray and dotted lines the maximum value that each MMSE subscale could reach: Max(TO) = 5, Max(OS) = 5, Max(Reg) = 3, Max(AC) = 5, Max(Rec) = 3, Max(N) = 2, Max(SR) = 1, Max(P) = 3, Max(W) = 1, Max(CE) = 1, Max(D) = 1. Abbreviations: CTRL = healthy control; PD = Parkinson’s disease patients; PSP = Progressive Supranuclear Palsy patients; TO = temporal orientation; OS = orientation in space, Reg = registration of three words; AC = attention and calculation; Rec = recall of three words; N = object naming; SR = sentence repetition; P = praxis; W = writing a sentence; CE = reading a sentence and close your eyes; D = copy a drawing.

**Figure 3.**Radar plots of the mean of the eleven MMSE subscales per cluster of Parkinson’s disease (PD) patients. In gray and dotted lines the maximum value that each MMSE subscale could reach: Max(TO) = 5, Max(OS) = 5, Max(Reg) = 3, Max(AC) = 5, Max(Rec) = 3, Max(N) = 2, Max(SR) = 1, Max(P) = 3, Max(W) = 1, Max(CE) = 1, Max(D) = 1. Abbreviations: TO = temporal orientation; OS = orientation in space, Reg = registration of three words; AC = attention and calculation; Rec = recall of three words; N = object naming; SR = sentence repetition; P = praxis; W = writing a sentence; CE = reading a sentence and close your eyes; D = copy a drawing.

**Figure 4.**Radar plots of the mean of the eleven MMSE subscales per cluster of Progressive Supranuclear Palsy (PSP) patients. In gray and dotted lines the maximum value that each MMSE subscale could reach: Max(TO) = 5, Max(OS) = 5, Max(Reg) = 3, Max(AC) = 5, Max(Rec) = 3, Max(N) = 2, Max(SR) = 1, Max(P) = 3, Max(W) = 1, Max(CE) = 1, Max(D) = 1. Abbreviations: TO = temporal orientation; OS = orientation in space, Reg = registration of three words; AC = attention and calculation; Rec = recall of three words; N = object naming; SR = sentence repetition; P = praxis; W = writing a sentence; CE = reading a sentence and close your eyes; D = copy a drawing.

**Figure 5.**Distribution of diagnoses among the four clusters found by affinity propagation. Cluster #1 consisted of 32 CTRL, 19 PD and 17 PSP. Cluster #2 consisted of 12 CTRL, 10 PD and 8 PSP. Cluster #3 consisted of 2 PD and 6 PSP. Cluster #4 consisted of 17 PD and 17 PSP. Abbreviations: CTRL = healthy control; PD = Parkinson’s disease patients; PSP = Progressive Supranuclear Palsy patients.

**Figure 6.**Radar plots of the mean of the eleven MMSE subscales per cluster. In gray and dotted lines the maximum value that each MMSE subscale could reach: Max(TO) = 5, Max(OS) = 5, Max(Reg) = 3, Max(AC) = 5, Max(Rec) = 3, Max(N) = 2, Max(SR) = 1, Max(P) = 3, Max(W) = 1, Max(CE) = 1, Max(D) = 1. Abbreviations: TO = temporal orientation; OS = orientation in space, Reg = registration of three words; AC = attention and calculation; Rec = recall of three words; N = object naming; SR = sentence repetition; P = praxis; W = writing a sentence; CE = reading a sentence and close your eyes; D = copy a drawing.

**Table 1.**Demographic and cognitive data of healthy controls (CTRL), Parkinson’s (PD) and Progressive Supranuclear Palsy (PSP) patients are reported as mean ± standard deviation. The mean and standard deviation of the residuals of the regression with age, sex and education as covariates, are also reported in round brackets.

CTRL (44) | PD (49) | PSP (48) | p-Value | Post-Hoc | |
---|---|---|---|---|---|

Age | 62.6 ± 11.5 | 66.7 ± 9.38 | 70.1 ± 8.32 | 0.002 ^{a} | CTRL < PSP ^{b} |

Female, n | 25 | 22 | 22 | N.S. ^{c} | |

Education | 12.5 ± 4.78 | 9.27 ± 4.74 | 7.38 ± 5.05 | <0.001 ^{a} | CTRL > PD ^{b}, PSP ^{b} |

Total MMSE | 29.2 ± 1.46 | 24.8 ± 5.08 | 20.8 ± 5.27 | <0.001 ^{d} | CTRL > PD ^{e}, PSP ^{e}; PD > PSP ^{e} |

TO | 4.93 ± 0.45 (0.241 ± 0.57) | 4.49 ± 1.04 (0.100 ± 1.010) | 3.85 ± 1.38 (−0.321 ± 1.250) | 0.01 ^{d} | CTRL > PSP ^{e} |

OS | 4.98 ± 0.15 (0.31 ± 0.567) | 4.29 ± 1.15 (−0.025 ± 0.992) | 3.81 ± 1.21 (−0.259 ± 0.999) | 0.003 ^{d} | CTRL > PSP ^{e} |

Reg | 3.00 ± 0 (0.008 ± 0.084) | 2.98 ± 0.143 (0.042 ± 0.142) | 2.85 ± 0.50 (−0.050 ± 0.485) | N.S.^{d} | N.A. |

AC | 4.80 ± 0.60 (0.854 ± 0.846) | 3.14 ± 1.90 (−0.167 ± 1.590) | 2.25 ± 1.77 (−0.616 ± 1.690) | <0.001 ^{d} | CTRL > PD ^{e}, PSP ^{e} |

Rec | 2.98 ± 0.15 (0.66 ± 0.39) | 1.96 ± 1.08 (−0.165 ± 0.970) | 1.52 ± 0.9 (−0.441 ± 1.010) | <0.001 ^{d} | CTRL > PD ^{e}, PSP ^{e} |

N | 1.98 ± 0.15 (−0.02 ± 0.14) | 2.00 ± 0 (0.030 ± 0.058) | 1.94 ± 0.32 (−0.015 ± 0.303) | N.S. ^{d} | N.A. |

SR | 1.00 ± 0 (0.10 ± 0.04) | 0.82 ± 0.39 (−0.055 ± 0.403) | 0.81 ± 0.39 (−0.038 ± 0.384) | 0.03 ^{d} | CTRL > PD ^{e} |

P | 2.93 ± 0.45 (−0.02 ± 0.43) | 2.98 ± 0.143 (0.088 ± 0.163) | 2.77 ± 0.55 (−0.074 ± 0.540) | N.S.^{d} | N.A. |

W | 0.91 ± 0.30 (0.16 ± 0.30) | 0.67 ± 0.47 (0.065 ± 0.394) | 0.29 ± 0.46 (−0.210 ± 0.465) | <0.001 ^{d} | CTRL > PSP ^{e};PD > PSP ^{e} |

CE | 0.82 ± 0.40 (−0.003 ± 0.410) | 0.84 ± 0.37 (0.123 ± 0.344) | 0.52 ± 0.50 (−0.120 ± 0.454) | 0.01 ^{d} | PD > PSP ^{e} |

D | 0.84 ± 0.37 (0.111 ± 0.270) | 0.65 ± 0.48 (0.083 ± 0.394) | 0.25 ± 0.44 (−0.185 ± 0.421) | <0.001 ^{d} | CTRL > PSP ^{e}; PD > PSP ^{e} |

^{a}ANOVA p-value, significant at p < 0.05.

^{b}Post-hoc of ANOVA corrected for multiple comparisons with Tukey’s, significant at p < 0.05.

^{c}Pairwise Chi-squared, significant at p < 0.05.

^{d}Analysis of covariance (ANCOVA) with age, sex and education in covariates, significant at p < 0.05.

^{e}Post-hoc of ANCOVA with age, sex and education in covariates, corrected for multiple comparisons with Tukey’s honest significant difference (HSD) test, significant at p < 0.05. Abbreviations: N.S. = not significant; N.A. = not applicable; CTRL = healthy control; PD = Parkinson’s disease patients; PSP= Progressive Supranuclear Palsy patients.

**Table 2.**Performance—evaluated with the Silhouette index—of the proposed algorithm by varying the preference value of the affinity propagation (AP) clustering.

Median Preference | Minimum Preference | |||
---|---|---|---|---|

Group | #Clusters | Silhouette Index | #Clusters | Silhouette Index |

CTRL | 8 | 0.302 | 1 | N.A. |

PD | 7 | 0.375 | 2 | 0.675 |

PSP | 6 | 0.387 | 2 | 0.677 |

CTRL + PD + PSP | 16 | 0.237 | 4 | 0.601 |

**Table 3.**Demographic and cognitive data of the four clusters found by affinity propagation. The mean and standard deviation of the residuals of the regression with age, sex and education as covariates, are also reported in round brackets.

Cluster #1 (68) | Cluster #2 (30) | Cluster #3 (8) | Cluster #4 (34) | p-Value | Post-Hoc | |
---|---|---|---|---|---|---|

Age | 62.4 ± 10.1 | 71.3 ± 9.94 | 70.4 ± 7.93 | 69.8 ± 7.64 | <0.001 ^{a} | 1 < 2,4 ^{b} |

Female, n | 33 | 14 | 5 | 17 | N.S. ^{c} | N.A. |

Education | 12.1 ± 5.23 | 6.60 ± 3.23 | 5.5 ± 2.67 | 8.59 ± 4.99 | <0.001 ^{a} | 1 > 2,3,4 ^{e} |

Total MMSE | 27.3 ± 3.93 | 27.7 ± 1.95 | 12.6 ± 2.92 | 20.1 ± 3.57 | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 1,3,4 ^{e}4 > 3 ^{e} |

TO | 4.68 ± 0.74 (0.026 ± 0.565) | 4.9 ± 0.55 (0.800 ± 0.550) | 1.25 ± 1.04 (−2.800 ± 0.842) | 4.18 ± 0.97 (−0.099 ± 0.888) | <0.001 ^{d} | 1 > 3 ^{e}2 > 1,3,4 ^{e}4 > 3 ^{e} |

OS | 4.71 ± 0.69 (0.067 ± 0.472) | 4.83 ± 0.46 (0.856 ± 0.487) | 2 ± 0.76 (−1.85 ± 0.43) | 3.74 ± 1.24 (−0.454 ± 1.02) | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 1,3,4 ^{e}4 > 3 ^{e} |

Reg | 2.93 ± 0.39 (−0.056 ± 0.372) | 3 ± 0 (0.108 ± 0.040) | 2.75 ± 0.46 (−0.139 ± 0.472) | 2.97 ± 0.17 (0.049 ± 0.162) | 0.016 ^{d} | 2 > 1 ^{e} |

AC | 4.21 ± 1.33 (0.345 ± 0.636) | 4.60 ± 0.72 (1.89 ± 0.619) | 0.50 ± 0.76 (−2.06 ± 1.11) | 1.21 ± 0.99 (−1.88 ± 0.767) | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 1,3,4 ^{e} |

Rec | 2.59 ± 0.69 (0.3 ± 0.637) | 2.50 ± 0.73 (0.021 ± 0.18) | 1.13 ± 0.83 (−0.775 ± 0.993) | 1.09 ± 0.93 (−0.937 ± 0.889) | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 3,4 ^{e} |

N | 2 ± 0 (0.007 ± 0.050) | 1.97 ± 0.183 (0.056 ± 0.302) | 1.75 ± 0.71 (−0.173 ± 0.675) | 1.97 ± 0.17 (0.008 ± 0.16) | N.S. ^{d} | N.A. |

SR | 0.89 ± 0.31 (0.005 ± 0.300) | 0.9 ± 0.30 (0.072 ± 0.532) | 0.50 ± 0.53 (−0.337 ± 0.535) | 0.88 ± 0.33 (0.019 ± 0.331) | 0.025 ^{d} | 3 < 1,2,4 ^{e} |

P | 2.94 ± 0.29 (−0.007 ± 0.268) | 2.90 ± 0.55 (0.201 ± 0.484) | 2.50 ± 0.76 (−0.312 ± 0.748) | 2.88 ± 0.41 (0.024 ± 0.416) | N.S.^{d} | N.A. |

W | 0.74 ± 0.41 (0.061 ± 0.316) | 0.67 ± 0.479 (0.218 ± 0.38) | 0.12 ± 0.35 (−0.333 ± 0.386) | 0.32 ± 0.47 (−0.22 ± 0.437) | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 3,4 ^{e} |

CE | 0.76 ± 0.43 (−0.039 ± 0.376) | 0.83 ± 0.379 (0.178 ± 0.45) | 0 ± 0 (−0.605 ± 0.075) | 0.71 ± 0.46 (0.029 ± 0.418) | <0.001 ^{d} | 1 > 3 ^{e}2 > 1,3 ^{e}4 > 3 ^{e} |

D | 0.79 ± 0.41 (0.070 ± 0.297) | 0.57 ± 0.50 (0.178 ± 0.450) | 0 ± 0 (−0.358 ± 0.212) | 0.26 ± 0.45 (−0.213 ± 0.402) | <0.001 ^{d} | 1 > 3,4 ^{e}2 > 3,4 ^{e} |

^{a}ANOVA p-value, significant at p < 0.05.

^{b}Post-hoc of ANOVA corrected for multiple comparisons with Tukey’s, significant at p < 0.05.

^{c}Pairwise Chi-squared, significant at p < 0.05.

^{d}ANCOVA with age, sex and education in covariates, significant at p < 0.05.

^{e}Post-Hoc of ANCOVA with age, sex and education in covariates, corrected for multiple comparisons with Tukey’s HSD test, significant at p < 0.05. Abbreviation: N.S. = not significant; N.A. = not applicable; CTRL = healthy control; PD = Parkinson’s disease patients; PSP = Progressive Supranuclear Palsy patients; TO = temporal orientation; OS = orientation in space, Reg = registration of three words; AC = attention and calculation; Rec = recall of three words; N = object naming; SR = sentence repetition; P = praxis; W = writing a sentence; CE = reading a sentence and close your eyes; D = copy a drawing.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Sarica, A.; Vaccaro, M.G.; Quattrone, A.; Quattrone, A.
A Novel Approach for Cognitive Clustering of Parkinsonisms through Affinity Propagation. *Algorithms* **2021**, *14*, 49.
https://doi.org/10.3390/a14020049

**AMA Style**

Sarica A, Vaccaro MG, Quattrone A, Quattrone A.
A Novel Approach for Cognitive Clustering of Parkinsonisms through Affinity Propagation. *Algorithms*. 2021; 14(2):49.
https://doi.org/10.3390/a14020049

**Chicago/Turabian Style**

Sarica, Alessia, Maria Grazia Vaccaro, Andrea Quattrone, and Aldo Quattrone.
2021. "A Novel Approach for Cognitive Clustering of Parkinsonisms through Affinity Propagation" *Algorithms* 14, no. 2: 49.
https://doi.org/10.3390/a14020049