Nonnegative matrix factorization for the identification of pressure ulcer risks from seating interface pressures in people with spinal cord injury

Abstract

The purpose of this study was to predict and visualize pressure ulcer risks by using a novel approach of extracting computational features from seating interface pressures in people with spinal cord injury (SCI). In conventional clinical practice, seating interface pressure assessments rely on descriptive statistics of pressure magnitude. In this study, rank-2 nonnegative matrix factorization (NMF) was applied to the seating interface pressure maps during loading and pressure-relieving conditions in 16 people with SCI. The NMF basis images were used for visual interpretation and computational prediction of pressure ulcer risks. The two NMF basis images encapsulated pressure concentration and pressure dispersion, respectively. The first basis converged on the ischial tuberosity under both seating conditions, whereas the second basis converged anterior to the ischial tuberosity during loading and converged on the coccyx during unloading. The classification yielded 81.25% overall accuracy. In general, higher ulceration risk was associated with higher and lower activations of the first and second bases, respectively. The NMF pipeline yielded promising performance. Basis visualization affirmed the importance of lower ischial pressure and higher distribution dispersion while also revealing that clinical practice may currently be underestimating the importance of coccygeal pressure in response to pressure-relieving activities. Graphical abstract.

Keywords: Dimensionality reduction; Feature extraction; Nonnegative matrix factorization; Pressure ulcers; Prevention; Spinal cord injury; Visualization.

Fig. 1

Nonnegative matrix factorization (NMF) basis images of the right buttock: NMF 1 under maximal loading (a), NMF 2 under maximal loading (b), NMF 1 under maximal unloading (c), and NMF 2 under maximal unloading (d)

Fig. 2

Mean pressure maps of the right buttock during maximal loading (a) and maximal unloading (b) superimposed with 90% contours of the nonnegative matrix factors NMF 1 (solid) and NMF 2 (dotted)

Fig. 3

Nonnegative matrix projection of participants with ulceration (filled) and nonulceration (unfilled) onto a stochastic gradient decision surface of predicted ulceration and predicted nonulceration; axis units in the projected space are arbitrary

Fig. 4

Mean pressure maps of the right buttock from participants with and without predicted wound history superimposed with 90% contours of nonnegative matrix basis vectors NMF 1 (solid) and NMF 2 (dotted): predicted ulceration under maximal loading (a), predicted nonulceration under maximal loading (b), predicted ulceration under maximal unloading (c), and predicted nonulceration under maximal unloading (d)

Fig. 5

Pressure maps of the right buttock from the lone false positive (i.e., predicted ulceration with actual nonulceration) during maximal loading (a) and maximal unloading (b) superimposed with 90% contours of nonnegative matrix basis vectors NMF 1 (solid) and NMF 2 (dotted)

Fig. 6

Pressure maps of the right buttock from a false negative (i.e., predicted nonulceration with actual ulceration) under maximal loading (a) and maximal unloading (b) superimposed with 90% contours of nonnegative matrix basis vectors NMF 1 (solid) and NMF 2 (dotted)

Fig. 7

Conceptual diagram of a computational pipeline for pressure ulcer detection using r-rank nonnegative matrix factorization of training data (a), projection and classification of the training data in the r-dimensional nonnegative feature space (b), and projection and classification of new data in the trained feature space (c)