Event Abstract

Back to Event

Predicting improvement in spelling after written naming/spelling treatment and tDCS in primary progressive aphasia: a machine learning approach

Vânia De Aguiar1*, Bronte Ficek2, Kimberly Webster3, Argye Hillis2, 4, 5 and Kyrana Tsapkini2, 5
  • 1 Trinity College Dublin, Department of clinical speech and language studies, Ireland
  • 2 Johns Hopkins Medicine, Department of neurology, United States
  • 3 Johns Hopkins Medicine, Department of otolaryngology, United States
  • 4 Johns Hopkins University, Department of physical medicine and rehabilitation, United States
  • 5 Johns Hopkins University, Department of cognitive science, United States

BACKGROUND In the last decades, a number of behavioral, pharmacological, and neuromodulatory interventions have been tested for people with Primary Progressive aphasia (PPA), with promising results in reducing behavioral consequences of FTLD (Croot et al., 2009; Nickels & Croot, 2016). Spelling therapy improves writing of trained words in this population (Tsapkini & Hillis, 2009) and results in generalization to untrained words when combined with transcranial Direct Current Stimulation (tDCS; Tsapkini et al., 2014). With these promising results in mind, it becomes important to consider which patients are more likely to improve with a particular treatment. METHOD Thirty-five individuals with PPA received written naming/spelling intervention combined with anodal tDCS over the left inferior frontal gyrus (IFG) or Sham, using a within-subjects, randomized, double-blind, cross-over design, with each intervention delivered over a period of 3 weeks and a washout period of 2 months between interventions. We report data concerning change in percentage of correctly spelled phonemes between the pre-treatment assessment, and the assessment immediately after treatment. Random forests (Breiman, 2002), a machine learning algorithm used in classification and regression, is used to identify predictors of improvement among demographic, clinical, and cognitive variables. The analyses followed the procedures outlined in de Aguiar et al. (2016). RESULTS Variables identified as informative predictors of percent change for correctly spelled graphemes are presented in Figure 1 (panels A and C). The best model evaluating combinations of these variables for trained words (C=0.4727) identified an interaction between sentence comprehension scores in object relatives and passives in the SOAP test (Love & Oster, 2002). PPA patients were likely to show greater change if pre-treatment scores in comprehension of object relatives were above 6.906. However, among those individuals with scores at or below 6.906 in object relatives, greater change was observed for patients with passive sentence comprehension scores equal or below 7.917. The latter were also lower in pre-treatment spelling accuracy than the former, for trained items. For change in spelling untrained words, an interaction between Digit Span Forward and Stimulation condition was identified in the best model (C=0.4784). PPA patients with pre-treatment Digit Span Forward scores at or below 3 were likely to show greater improvement than those with scores above this level. The former individuals also showed significantly lower accuracy in spelling words and nonwords. Thus, this is a subgroup of patients with greater spelling impairment, and therefore more likely to benefit from treatment. In those with Digit Span Forward above 3, greater improvement was observed when they received anodal tDCS than when they received Sham (Figure 1, panel D). DISCUSSION AND CONCLUSION A detailed analysis of all informative predictors can highlight some potential mechanisms of improvement behind treatment-related changes in performance in individuals with PPA, and selectivity in the response to tDCS treatment both for a particular outcome (untrained words) and for patients with particular profiles (good verbal short-term memory). However, the low accuracy of the models (C=0.4727 and 0.4784) reflect the need to consider other potential predictors, and to confirm findings with a larger database.

Figure 1
Image

Acknowledgements

We are grateful to the participants in this project and for the support from the Science of Learning Institute of Johns Hopkins University and NIH/NIDCD R01 DC014475 to KT.

References

Breiman, L. (2002). Manual on Setting Up, using, and Understanding Random Forests v3. 1. Statistics Department University of California Berkeley, CA, USA. Available online at: https://www.stat.berkeley.edu/~breiman/Using_random_forests_v4.0.pdf
Croot, K., Nickels, L., Laurence, F., & Manning, M. (2009). Impairment‐and activity/participation‐directed interventions in progressive language impairment: Clinical and theoretical issues. Aphasiology, 23(2), 125-160.
de Aguiar, V., Bastiaanse, R., & Miceli, G. (2016). Improving Production of Treated and Untreated Verbs in Aphasia: A Meta-Analysis. Frontiers in Human Neuroscience, 10.
Love, T., & Oster, E. (2002). On the categorization of aphasic typologies: The SOAP (a test of syntactic complexity). Journal of Psycholinguistic Research, 31(5), 503-529.
Nickels, L., & Croot, K. (2014). Understanding and living with primary progressive aphasia: Current progress and challenges for the future. Aphasiology, 28(8-9), 885-899.
Tsapkini, K., Frangakis, C., Gomez, Y., Davis, C., & Hillis, A. E. (2014). Augmentation of spelling therapy with transcranial direct current stimulation in primary progressive aphasia: Preliminary results and challenges. Aphasiology, 28(8-9), 1112-1130.
Tsapkini, K., & Hillis, A. E. (2013). Spelling intervention in post-stroke aphasia and primary progressive aphasia. Behavioural neurology, 26(1, 2), 55-66.

Keywords: PPA, tDCS, intervention, spelling, Writing, language rehabilitation, Random forests, machine learning

Conference: Academy of Aphasia 55th Annual Meeting , Baltimore, United States, 5 Nov - 7 Nov, 2017.

Presentation Type: poster or oral

Topic: General Submission

Citation: De Aguiar V, Ficek B, Webster K, Hillis A and Tsapkini K (2019). Predicting improvement in spelling after written naming/spelling treatment and tDCS in primary progressive aphasia: a machine learning approach. Conference Abstract: Academy of Aphasia 55th Annual Meeting . doi: 10.3389/conf.fnhum.2017.223.00107

Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.

The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.

Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.

For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.

Received: 02 May 2017; Published Online: 25 Jan 2019.

* Correspondence: PhD. Vânia De Aguiar, Trinity College Dublin, Department of clinical speech and language studies, Dublin, Ireland, deaguiar@jhmi.edu