Interpreting PET scans by structured patient data: a data mining case study in dementia research

Abstract

One of the goals of medical research in the area of dementia is to correlate images of the brain with clinical tests. Our approach is to start with the images and explain the differences and commonalities in terms of the other variables. First, we cluster Positron emission tomography (PET) scans of patients to form groups sharing similar features in brain metabolism. To the best of our knowledge, it is the first time ever that clustering is applied to whole PET scans. Second, we explain the clusters by relating them to non-image variables. To do so, we employ RSD, an algorithm for relational subgroup discovery, with the cluster membership of patients as target variable. Our results enable interesting interpretations of differences in brain metabolism in terms of demographic and clinical variables. The approach was implemented and tested on an exceptionally large data collection of patients with different types of dementia. It comprises 10 GB of image data from 454 PET scans, and 42 variables from psychological and demographical data organized in 11 relations of a relational database. We believe that explaining medical images in terms of other variables (patient records, demographic information, etc.) is a challenging new and rewarding area for data mining research.

Mining bridging rules between conceptual clusters

“…In Table 2, in the first and second cases (minsim = 0.6, minentro = 6 and minsim = 0.75, minentro = 6), all the bridging rules belong to the clusters mammal and Table 2 The results generated by the joint-entropy-based strategy (25,82), (3,77), (3,20), (106, 125), (61, 67), (24,125), (115, 125), (72, 125), mininfo = 1.5, p = 1 (122, 256), (19,33), (19,67), (25,100), (25,73) (19,32), (19,67), (19,33), (3,123), (27,78) Table 4 The results generated by the joint-entropy-based strategy Thresholds Bridging rules minsim = 0.6, minentro = 6.2 (18,47), (6,44), (26,44), (18,44), (30,44), (18,42), (22,42), (17,44), (2,41), (28,38), (29,42), (1...…”

“…An outlier [1][2][3]25] is defined as a data object that is grossly different from the remaining set of data objects. In many applications [2][3][4][5][6][7][8][9][10], finding outliers can be very problematical, because one person's noise can be another person's signal, and so various algorithms [11][12][13][14][15][16][27][28][29] have been used for detecting outliers from different angels. In this paper we define a new kind of outlier: a bridging rule, whose antecedent and action belong to different conceptual clusters, that represents certain interactions between clusters.…”

Subgroup Discovery Algorithms: A Survey and Empirical Evaluation

“…[41] employs an SD approach SD4TS (Subgroup Discovery for Test Selection) for breast cancer diagnosis. Some other studies [42][43] have detected the groups of gene that are highly correlated with the class of lymphoblastic leukemia. They have employed the RSD algorithm for inducing subgroups that characterize the genes in terms of the knowledge extracted from gene ontologies and interactions.…”

Identifying consistent statements about numerical data with dispersion-corrected subgroup discovery

“…Subgroup discovery is a well-established KDD technique (Klösgen (1996); Friedman and Fisher (1999); Bay and Pazzani (2001); see Atzmueller (2015) for a recent survey) with applications, e.g., in Medicine (Schmidt et al, 2010), Social Science (Grosskreutz et al, (a) optimizing coverage times median shift; σ0(·) ≡ (a(·) ≥ 6) ∧ (c2(·) > 0) ∧ (c6(·) < v. high); subgroup median 0.22, subgroup error 0.081 (b) optimizing dispersion-corrected variant; σ1(·) ≡ (a(·) ∈ [8, 12])∧(c2(·) > low)∧(c6(·) < v. high) ∧ (r(·) > v. low); subgroup median 0.23, subgroup error 0.028 Figure 1. To gain an understanding of the contribution of long-range van der Waals interactions (y-axis; above) to the total energy (x-axis; above) of gas-phase gold nanoclusters, subgroup discovery is used to analyze a dataset of such clusters simulated ab initio by density functional theory (Goldsmith et al, 2017); available features describe nanocluster geometry and contain, e.g., number of atoms a, fraction of atoms with i bonds c i , and radius of gyration r. Here, similar to other scientific scenarios, a subgroup constitutes a useful piece of knowledge if it conveys a statement about a remarkable amount of van der Waals energy (captured by the group's central tendency) with high consistency (captured by the group's dispersion/error); optimal selector σ 0 with standard objective has high error and contains a large fraction of gold nanoclusters with a target value below the global median (0.13) (a); this is not the case for selector σ 1 discovered through dispersion-corrected objective (b), which therefore can be more consistently stated to describe gold nanoclusters with high van der Waals energy.…”