Skip to main content

Advertisement

Fig. 7 | BioMedical Engineering OnLine

Fig. 7

From: Microneurography as a tool to develop decoding algorithms for peripheral neuro-controlled hand prostheses

Fig. 7

Decoding. a, b Confusion matrices of the requested versus decoded action (rest included) representing the performance of the different decoders when predicting forces (left) and velocities (right). They were obtained by the sum of the single normalized matrices computed for each subject separately. f1,2,3,4 = force 1,2,3,4, v1,2,3 = velocity 1,2,3. c Overall performance of all the subjects for custom force and velocity decoding applied to isotonic and isokinetic tasks. Chance level is shown in red. d, e Requested action, computed features, and action decoded using our algorithm, for isotonic (top) and isokinetic (bottom) tasks extracted respectively from Subj. 1 and 6. f Normalized performance of the used decoders applied to all subjects for isotonic and isokinetic tasks. CC custom classifier (i.e. the classifier we developed), CF custom features (i.e. the features extracted by our motoneuron behavior analysis), SU single-unit, MU multi-unit. In cf colored circles represent results from the different subjects. Force and velocity decoders are trained and tested on sets of 12 and 9 repetitions per subject, respectively. Kruskal–Wallis tests are executed on data in c, f and Tukey–Kramer correction is also performed on data in f. ** means p < 0.05

Back to article page