Microstructural properties in subacute aphasia: Concurrent and prospective relationships underpinning recovery

Abstract

Background:

Approximately one in three people who have a stroke will experience aphasia, or impairment of language, as a devastating consequence of the subsequent damage to the language network of the brain. Few investigations have been carried out to examine the relationship between microstructural white matter integrity and subacute post-stroke linguistic performance or the relationship between microstructural integrity and the recovery of language function.

Aims:

We examined two key questions regarding the relationship between microstructural integrity and recovery: 1) How does subacute language performance as measured both in single words (i.e., naming pictures of objects) and in discourse (the amount of unique content (content units, CU) included when describing a picture and the efficiency with which that content in conveyed) relate the integrity of key white matter regions of interest in the language network? and 2) Does the integrity of these regions before treatment predict the improvement or resolution of linguistic symptoms following treatment both immediately and chronically?

Methods:

58 participants within the first 3 months of stroke were enrolled in a randomized, single-center, double-blind, sham-controlled, study of anodal transcranial direct current stimulation combined with a computer-delivered speech and language naming therapy for subacute aphasia (ClinicalTrials.gov NCT02674490) and were asked to complete magnetic resonance imaging at enrollment. Microstructural integrity was evaluated using diffusion tensor imaging processed with atlas-based segmentation. Pearson correlation coefficients were calculated between mean diffusivity of each of the regions of interest important for language and performance in picture naming of nouns, and content and efficiency of language when describing an illustrated scene.

Results:

22 participants received diffusion tensor imaging. All participants had aphasia at baseline and recovered language to some degree over the course of the study. Picture naming accuracy significant correlated with poorer mean diffusivity in the left posterior inferior temporal gyrus. Description efficiency was significantly positively correlated with mean diffusivity in the posterior superior, middle, and inferior temporal gyri, as well as the angular gyrus. Description content did not correlate with mean diffusivity of any region of interest. Recovery of naming performance was predicted by days since stroke and baseline microstructural integrity of the left posterior medial temporal gyrus, arcuate fasciculus, and superior longitudinal fasciculus. Recovery of discourse efficiency was significantly predicted by the same model.

Conclusions:

This study demonstrates association between picture naming and discourse and microstructural integrity of the key regions in the language network for patients with subacute post-stroke aphasia. Baseline microstructural integrity significantly predicts language recovery.

Data access statement:

Data will be available from the Johns Hopkins Data Archive upon publication.

Introduction

Approximately one in three people who have a stroke will experience aphasia, or impairment of language, as a devastating consequence of the subsequent damage to the language network of the brain.^{1, 2} While abundant research has been successful in defining structure-function relationships between areas of damage and specific language deficits (e.g., production versus comprehension or fluent versus non-fluent syntax), the relationship between areas of ischemic brain damage and impairment is imperfect. Often, seemingly spared areas appear to be associated with persistent dysfunction.^{3, 4} However, fewer investigations have been carried out to examine the relationship between microstructural white matter integrity and subacute post-stroke linguistic performance or the relationship between microstructural integrity and the recovery of language function. What has emerged with increasing clarity is that the recovery of language appears dependent on bilateral white matter changes, with levels of relative dominance shifting over time after stroke.^5–8

Key regions associated with language function include the left posterior superior (pSTG), middle (pMTG), and inferior temporal gyri (pITG), angular gyrus (AG), arcuate fasciculus (AF), inferior fronto-occipital fasciculus (IFOF), and inferior (ILF) and superior longitudinal fasciculi (SLF).^9–12 Diffusion tensor imaging (DTI) uses magnetic resonance imaging (MRI) technology to calculate scalars of water molecule diffusion from atlas-segmented white matter tracts, and has been employed to better understand the relationship between microstructural and functional change in language in stroke recovery.^{13, 14} Mean diffusivity (MD) measures the average of radial and axial water displacement associated with cellular damage. Higher MD values are conventionally shown to be associated with diminished microstructural integrity.

Auditory comprehension deficits at the single word level (i.e., difficulties identifying the correct picture of a noun or verb from a spoken cue) have been associated with impaired tract integrity in the IFOF and ILF.¹⁵ Production deficits when asked to name pictures objects and actions has been associated with decreased integrity in the AF.¹⁵ In contrast, discourse production deficits (in producing unambiguous linguistic relationships such as reference across sentences) have been associated with changes to not only the AF, but also the IFOF and ILF.¹⁴ Aphasia severity on the whole, as defined by the summary quotient score on the Western Aphasia Battery,¹⁶ has been associated with modules containing the STG, MTG, and ITG in a connectome analysis.¹² In the context of treatment-related microstructural change, ILF, SLF, and AF changes have been observed associated with decreasing overall severity,¹⁷ with changes in the ILF and AF¹⁸ specifically associated with a treatment response in naming pictures.

Aims

Here, we present the findings from a study treating subacute post-stroke aphasia using speech language treatment (SLT) with or without augmentation via the addition of transcranial direct current stimulation (tDCS) in which we examine two key questions regarding the relationship between microstructural integrity and recovery: 1) How does subacute language performance as measured both in single words (i.e., naming pictures of objects) and in discourse (the amount of unique content (content units, CU) included when describing a picture and the efficiency with which that content in conveyed) relate the integrity of key white matter regions of interest in the language network? and 2) Does the integrity of these regions before treatment predict the improvement or resolution of linguistic symptoms following treatment both immediately and chronically?

Methods

Recruitment

Participants (N=58) were enrolled in a randomized, single-center, double-blind, sham-controlled, study of anodal transcranial direct current stimulation combined with a computer-delivered speech and language therapy for subacute aphasia, previously reported on in Stockbridge et al.¹⁹ All procedures were approved by the Johns Hopkins Medicine Institutional Review Board (IRB00089018) and all participants or their legally authorized representatives provided written consent. The study was registered with ClinicalTrials.gov (NCT02674490). The study was completed at Johns Hopkins Hospital and Johns Hopkins Bayview Medical Center in Baltimore, Maryland, between September 16, 2016 and October 4, 2021 (final follow-up evaluation March 29, 2022). Included individuals were right-handed, adult English speakers within three months of ischemic left hemisphere stroke who were diagnosed with aphasia using the Western Aphasia Battery-Revised (WAB-R).¹⁶ Individuals with previous neurological or psychiatric disease, seizures, brain surgery, metal in the head, uncorrected visual or hearing loss, scalp sensitivity, and certain medications were excluded.

Treatment

Patients completed 15 45-minute sessions of computer-delivered naming treatment, identical to the treatment used in Fridriksson et al.²⁰ with a speech-language pathologist. During naming treatment, patients were shown an on-screen picture of an object (e.g., cat) for two seconds then viewed a video of a female speaker saying a word that was either the name of the object (e.g., “cat”) or a semantic foil (“dog”), phonological foil (“hat”), or unrelated foil (“apple”). Patients were asked to identify whether the picture and the audio/video presentation of the word matched and received immediate visual feedback. Half of all patients (N=58) were randomized to receive anodal tDCS during naming treatment while the remainder received a sham treatment; however, treatment assignment was not able to be considered in analysis due to small sample sizes. There were no significant differences between tDCS and sham groups in recovery of naming on the PNT as measured one, five, or twenty weeks after treatment.¹⁹

Evaluation

Participants completed the 175-item color Philadelphia Naming Test²¹ (PNT; naming pictures of objects). In order to arrive at stable measures of naming that demonstrated high fidelity, the items were administered twice on consecutive days for each timepoint. The task was recorded and scored remotely by individuals blinded to the patient’s clinical profile, and then scores on each administration were averaged to reduce variability. Patients also were asked to describe the Cookie Theft picture on the NIH Stroke Scale²² (NIHSS, a measure of overall stroke severity), which was analyzed for Content Units (CU) and Syllables/CU.^{23, 24} CU are conceptual targets mentioned by healthy controls when describing the picture and provide a sensitive measure of the content of communication.^{24, 25} The Cookie Theft picture includes 30 CUs on the left and 23 on the right combined into a single total, providing comparable opportunities for the inclusion of information in the picture description regardless of attentional neglect. Syllables/CU is measured by the total number of syllables (including nonwords such as “uh” and perseverations) in the description, divided by CU, such that lower syllable/CU indicates more efficient discourse. For individuals who produce no CU in their sample, syllables are divided by 0.1. One, five, and twenty weeks after the final treatment session, patients were re-evaluated.

MRI acquisition and preprocessing

The MRI protocol included high-resolution T1weighted, T2weighted, Fluid Attenuated Inversion Recovery (FLAIR), and Diffusion-weighted imaging (DWI), and data were acquired on a Philips 3T scanner with a SENSE receive head coil. The details of MRI acquisition protocol were described elsewhere;²⁶ 32 direction DWI data were acquired axially with no gaps. TR/TE:7013/71, FOV=212×212 mm, the b-value=700s mm2, slice thickness=2.2 mm/in-plane resolution=0.82 mm. We converted DWI data files into nifti format by using dcm2nii (nitrc.org/plugins/mwiki/index.php/dcm2nii:MainPage). Once nifti files were generated, we used DSIstudio (dsistudio.labsolver.org) for diffusion analysis. The source images were inspected for significant movement artifacts, and the eddy motion correction was performed. We then resampled diffusion-weighted images at 1 mm isotropic. Additionally, the b-table was checked by an automatic quality control routine to ensure its accuracy.

Lesion Load Segmentation

For quality assurance purposes, lesion load included white matter (WM) hyperintensities and infarct volumes and were quantified manually in MRIcron software (nitrc.org/projects/mricron) by an experienced neurologist (ZK) by using high-resolution T2-weighted and FLAIR and T1 weighted images as also described elsewhere.¹⁴ Lesions identified were also inspected in the b0 map of DWI images during atlas-based segmentation.

Atlas Based Segmentation

We obtained DTI maps such as fractional anisotropy (FA; μ ± σ), mean (MD), radial (RD) and axial diffusivity maps (10−3 mm2 s−1) by processing data in DSI studio. Once the maps are generated, we used embedded maps Brainnectome, FreeSurfer, Human Connectome Project tractography (HCP842) maps to generate regions left posterior pSTG (merged regions STG L 6_2, 6_4; Brainnectome map), pMTG (MTG 4_3; Brainnectome map), pITG (merged regions ITG 7_2, 7_5; Brainnectome map), AG (IPL 6_5; Brainnectome map), AF (HCP842 map), IFOF (HCP842 map), ILF (HCP842 map), and SLF (HCP842 map). Cerebrospinal fluid contamination and ischemic infarcts were removed from the regions by using the lower FA threshold of 0.15 and the upper threshold RDs of 1.5-×-10−3 mm2 sec−1 (Figure 1).^{26, 27}

Figure 1.

The illustration of the interplay between the lesion and the superior longitudinal and arcuate fasciculi in two subjects. Of note, due to the thresholding, there was no overlap between lesions and regions of interest.

Regions were then edited with neuroanatomy guidance on slice-by-slice basis on DTI maps by moving the volume of interest in 3D space, removing neighboring tissues with smoothing and shaving functions. Our group previously observed that in subacute period, MD values were more sensitive in reflecting changes related to microstructural integrity of the tracts than FA or other diffusivity values.^{26, 28} Thus, in this study, we only included the MD values of left hemispheric structures.

Statistical Analyses

In order to examine the relationships between baseline measures and microstructural integrity, Pearson correlation coefficients were calculated between MD of each of the regions (left pSTG, pMTG, pITG, AG, AF, IFOF, ILF, and SLF) and available measures of baseline performance: PNT, CU, and Syllables/CU.

In order to determine the predictive value of baseline microstructural integrity on improvement of language skills, a series of regressions were calculated controlling for age and the number of days between stroke and the start of the study (days post onset). For each behavioral outcome measure (PNT, CU, and Syllables/CU), a separate regression was calculated to predict change in that variable sub-acutely (one week after treatment) and chronically (combining performance five- and twenty-weeks post-treatment to maximize the data available for analysis). To address study-wide error α=0.05 across all planned regressions, a two-tailed sequentially rejective multiple test procedure was used.²⁹

Results

Participant description

Twenty-two participants completed diffusion tensor imaging, summarized in Tables 1–3. Aphasia types varied across the sample: one global, five Broca’s, three Wernicke’s, three Transcortical Motor, one Transcortical Sensory, one Conduction, and eight anomic. Participants improved in language over the course of the study, as measured by increasing numbers of items correct on the PNT (Greenhouse-Geisser corrected repeated measures analysis of variance: F(1.4, 28.4) = 22.6, p < 0.001, η_P² = 0.53). Although trends toward improvement were present, there were no significant differences in quantities of content conveyed (F(1.2, 22.1) = 0.53, p = 0.51) or syllables needed to convey content when describing the Cookie Theft picture (F(1.0, 18.2) = 1.9, p = 0.19).

Table 1:

Baseline demographics (mean ± standard deviation)

ID	Age	M:F	Edu	DPO	NIH SS	AQ	Aphasia	Stim	Lesion Location	Vol
1	78	M	20	69	2	50	TCM	tDCS	Left temporal pole, anterior ITG/MTG, lentiform nucleus and caudate	82.28
2	80	M	12	59	3	71.6	Anomia	tDCS	Left IFG/MFG	25.3
3	65	F	20	55	2	77.6	Anomia	tDCS	Left posterior ITG, MTG, STG	34.07
4	67	F	16	52	3	46.9	Wernicke’s	tDCS	Left IFG	6.62
5	83	F	13	89	6	59.9	Wernicke’s	Sham	Left pSTG, SPG, cerebellum	4.38
6	30	M	12	82	12	71.2	Anomia	Sham	Left caudate, lentiform, thalamus, left pSTG, IPG	32.68
7	72	F	16	14	5	15.9	Broca’s	Sham	Left pSTG, fusiform gyrus, external capsule	22.54
8	69	F	13	75	3	52.5	TCM	Sham
9	74	M	16	27	2	48.7	Broca’s	tDCS	Left pSTG and external capsule	36.25
10	70	F	12	78	1	92.8	Anomia	tDCS	Left IFG, pSTG, right cerebellum	15.67
11	74	M	20	24	3	60.9	TCS	tDCS	Left IFG, corona radiata	2.99
12	79	M	20	94	9	48.6	Broca’s	Sham	Left IFG, IPG	24.9
13	61	M	12	48	5	82.3	TCM	Sham	Left pSTG, IPG, SPG, SFG	75.48
14	57	M	12	16	7	46.2	Broca’s	Sham	Left IFG, MFG, pSTG	29.94
15	84	M	16	9	3	27.2	Global	tDCS	Left pSTG	0.94
16	75	M	12	83	2	87	Conduction	tDCS	Left pSTG	3.64
17	73	F	18	47	2	87.4	Anomia	tDCS	Left pSTG	5.36
18	47	M	16	41	4	85.7	Anomia	Sham	Left calcarine gyrus, fusiform gyrus	16.07
19	69	F	16	20	2	93.3	Anomia	tDCS	Left SPG	30.78
20	55	M	18	100	12	30.1	Wernicke’s	tDCS	Left pSTG, IPG, SPG	60.4
21	86	F	12	60	4	88.8	Anomia	tDCS	Left IFG	0.44
22	67	F	12	45	7	47.9	Broca’s	tDCS	Left precentral gyrus, IPG, STG, calcarine	121.49
	69±13	12:10	15±3	54±28	5±3	62±23		14:8

Edu: Education. DPO: Days post onset. AQ: Aphasia Quotient on the Western Aphasia Battery. Aphasia: Aphasia subtype. TCM[S]: Trans-cortical motor [sensory]. Participants were randomized to receive either tDCS or sham during treatment. Vol: Infarct volume (mL).

Table 3:

Performance at baseline and follow-up assessments (mean ± standard deviation)

	PNT*				Content Units				Syllables/Content Unit
ID	Baseline	1 wk	5 wks	20 wks	Baseline	1 wk	5 wks	20 wks	Baseline	1 wk	5 wks	20 wks
1	78.5	113.5	99	121	2	6	--	3	20	11.83	--	25.67
2	98	107	114	112.5	6	7	16	9	4.8	5	5.69	6.56
3	135.5	149.5	151.5	157.5	12	12	16	15	8.67	16.17	8.9	8.7
4	29.5	62	73	82	3	6	9	14	25.3	26.17	29.44	26.28
5	38.5	51.5	52	63	6	12	7	9	17.5	13.91	42.9	20.8
6	127	--	147	145.5	6	--	10	7	7.5	--	3.5	7.3
7	0	4	6	15	2	9	8	10	45.2	55.4	61.13	23.2
8	125.5	124	125.5	142	7	6	9	6	8.9	9.3	7.8	11.83
9	49.5	93.5	96	104	5	15	13	14	61.2	19.9	13.7	15
10	162	167.5	163	165	13	15	11	14	8.08	5.2	7.73	5.85
11	73	108	108	117	1	10	9	7	47	6.5	10.7	6.28
12	20.5	25.5	30	15.5	--	--	--	--	--	--	--	--
13	161.5	167.5	163	161	65	12	13	12	11.7	4.7	6	5.75
14	56.5	106	104	111.5	9	6	8	12	10.33	7.83	12.5	12.16
15	111	140	144.5	141	6	12	16	15	9.7	8.33	4.19	6.8
16	150	158.5	159.5	159	14	11	10	14	2.57	5	5.7	4.35
17	136	165	170	167.5	18	19	21	27	47.9	23.8	21.3	11.5
18	42	98.5	110.5	132.5	0	3	4	14	N/A	25	41.5	14.5
19	149	157	156.5	158.5	15	16	13	12	8.13	12.5	10.23	11.67
20	6.5	4.5	8	11	3	5	5	9	84	41	41	30.33
21	150.5	154.5	158	160.5	17	14	16	24	4.4	14	9	6.54
22	26.5	33	39.5	68	1	4	5	8	70	22.5	17.6	10.63
	86±55	104±55	106±53	113±51	11±14	10±5	11±5	13±5	55±127	17±13	19±16	13±8

^**p < 0.001 in Greenhouse-Geisser corrected repeated measures analysis of variance

Relationship between baseline performance and microstructural integrity

Pearson correlations are summarized in Figure 2. Picture naming performance demonstrated a significant but weak correlation with lower mean diffusivity in the left pITG. Discourse measures at baseline did not correlate with mean diffusivity of any region of interest.

Fig. 2.

Correlations between baseline performance and mean diffusivity

*p < 0.05, **p < 0.01. Darker shaded regions are associated with stronger relationships between variables. Days: Days post onset.

Predicting recovery from baseline microstructural integrity

Recovery following language treatment was significantly predicted by baseline mean diffusivity in multiple key regions of interest (Table 4). Short-term maintenance of treatment effects (the difference between baseline performance and performance measured one week after language treatment) on picture naming of objects was significantly predicted by days since stroke and microstructural integrity of the pMTG, AF, and SLF, with the full model accounting for 78% of the variance in PNT recovery. Greater MD in the AF was associated with better performance in picture naming, while lower diffusivity of the pMTG and SLF were predictive of greater improvement. The model predicting change in content on the Cookie Theft was not significant. However, change in efficiency on the same task was predicted significantly by the model, explaining 90% of the variance. In addition to age, lower diffusivity in the pITG, AG, AF, and SLF also significantly predicted better recovery of efficiency.

Table 4:

Prediction of recovery from microstructural integrity of each region

	Short-term maintenance (1-week after therapy)			Long-term maintenance (5- and 20-weeks after therapy)
	PNT (N = 21)	CU (N = 20)	Syllables/CU (N = 20)	PNT (N = 22)	Syllables/CU (N = 20)
Constant	70.1	−31.0	287.6	82.4	25.6
Age	−0.26	0.61	0.78*	−0.24	0.28
Days	−0.83*	−0.77	0.31	−0.97**	0.49
pSTG	0.02	−0.08	−0.32	0.02	−0.51
pMTG	−0.86*	−0.02	0.55	−0.87*	0.38
pITG	0.47	0.81	−0.52*	0.57*	−0.73*
AG	0.12	−0.39	0.97*	0.18	0.46
AF	1.24*	1.06	−1.80*	1.42*	−2.06*
IFOF	0.22	−0.31	0.37	0.15	0.89*
ILF	0.11	0.19	−0.34	0.13	−0.72*
SLF	−1.50*	−1.39	1.33*	−1.75**	1.56*
F(df)	3.58(10, 10)	1.05(10, 9)	8.19(10, 9)	6.36(10, 11)	3.85(10, 10)
R2	0.78	0.54	0.90	0.72	0.79
p	0.03	0.47	0.002	0.003	0.02

^*p<0.05,

^**p<0.001

Variables of interest reflect change in scores from baseline, not raw scores at each timepoint. Standardized coefficients are reported for variables of interest. Days: Days from stroke to start of therapy.

Models predicting long-term improvement (the average of performance measured 5- and 20-weeks from the end of treatment) from microstructural integrity was significant for both measures considered, PNT and informational efficiency. PNT improvement was significantly predicted by lower diffusivity in the pMTG and SLF and higher diffusivity in the pITG and AF. The pattern of significant predictors was generally similar to what was observed short term. Lower diffusivity in the pITG, AF, and ILF and higher diffusivity in the IFOF and SLF predicted better recovery of efficiency. The number of days between stroke and treatment was a significant and influential factor on PNT improvement, but not change in informational efficiency.

Discussion

In the present study, we examined two key questions regarding subacute aphasia. The first was whether language at the level of either single words or discourse related to the microstructural integrity of key regions of interest: the left pSTG, pMTG, pITG, AG, AF, IFOF, ILF, and SLF. Picture naming correlated with integrity of the left pITG, consistent with previous reports regarding the importance of pITG for naming.^30–32 Discourse measures were not correlated significantly with integrity of any structures when examined at baseline. This perhaps is unsurprising, as discourse likely depends on a number of different cognitive functions, including word-finding, syntax, working memory, and lexical semantics, both within and beyond the language network.

The second question we examined was whether microstructural integrity of the regions of interest predicted either short- or long-term improvement in single word or discourse level language. Along with days since stroke and age, microstructural integrity predicted 78% of the variance in short-term picture naming recovery and 90% of the variance in short-term recovery of efficiency. Days and MDs of the pMTG, AF, and SLF contributed significantly to the model predicting naming while age and MDs of the pITG, AG, AF, and SLF significantly contributed to the prediction of efficiency improvement. Short-term recovery of content was not predicted significantly by these factors. Prediction of long-term recovery of picture naming relied on similar factors with similar success. In addition to days since stroke, MDs of the pMTG, pITG, AF, and SLF also contributed significantly to the prediction. Prediction of long term-recovery of discourse efficiency varied slightly from prediction of recovery in the short-term. Prediction of long-term improvement was not contributed to significantly by the integrity of the AG, but integrity of the IFOF and ILF reached significance. Days since stroke was no longer a significant contributor to the model. Change in content provided on the Cookie Theft picture description was very small with high variability across the study period, perhaps contributing to the absence of correlations between this measure and baseline MD of any structure of interest.

One of the major challenges in in understanding and predicting post stroke recovery is the amount of heterogeneity observed across the individuals. Although post-stroke motor recovery field has validated prediction tools,³³ the field of post-stroke language recovery is lacking consistent, validated biomarkers to predict the natural recovery of each individual. Similarly, we do not have widely accepted biomarkers to predict treatment response to experimental language therapies.³⁴ MRI based neuroimaging is a non-invasive but a powerful tool that can provide biomarkers for prediction of outcome and of treatment response in ischemic stroke.³⁵ Additionally, DTI previously was shown to reveal dysfunction in the language network and identify recovery predictors beyond the stroke lesions.^{26, 36} In our study, we attempted to characterize baseline DTI characteristics of key language regions and how they predict the language recovery after a computer delivered lexical-semantic therapy alone or combined with tDCS. Studies like ours have potential implications in allocating patients with post-stroke language deficits into tailored behavioral therapies in order to maximize their recovery potential.

Prior studies have explored the relationship between microstructural integrity and picture naming with similar results. We have shown MD of left AF and SLF predicted picture naming recovery, and this is in line with previous studies showing SLF and AF associated with poorer picture naming functions and recovery.^{37, 38} Although it is not very clear to us why increased MD of AF at baseline predicted a better naming recovery whereas increased MD of SLF a poorer recovery, depending upon the lesion location, diffusivity metrics in the perilesional areas can evolve in the subacute period differently than chronic period, for which we have the most data. Our data also supports a ongoing evolution of diffusivity metrics with an uptrend in MD values Further studies including patients in the subacute period should further elaborate on this.

pMTG previously has been shown to play role in semantic retrieval.³⁹ Similarly, in our study, pMTG was found to predict picture naming recovery. Additionally, recent work in primary progressive aphasia also found a significant relationship between treatment-supported improvement in naming and baseline characteristics of the pITG.⁴⁰ Our correlation analyses that show positive correlation between microstructural integrity and picture naming performance are in line with the existing literature.

In contrast, relatively few teams have examined the relationship between microstructural integrity and discourse measures after stroke. Given the limited available sample size and relative stability of content inclusion among individuals within the sample, whose language treatment did not target discourse skills directly, it seems possible that this study was underpowered to observe differences in discourse content. Null findings also were reported previously by our group when examining tracts related to recovery of discourse content in a different design.¹⁴ However, to our knowledge, no prior study has examined the relationship between microstructural integrity and informational efficiency in discourse – either contemporaneously or predictively. We found that recovery was strongly influenced by integrity of a large array of structures within the language network, consistent with the diverse skills required for discourse to be successful.

Limitations

One of the main limitations is small sample size with heterogenous aphasia subtypes. Another limitation is we have analyzed together patients who received active tDCS and those who received sham tDCS due to small sample size. However, this limitation is mitigated by the absence of a significant effect of tDCS on change in naming measured with the PNT. Our study also utilized a conventional DTI rather than advanced diffusion techniques (i.e., high resolution, multi-shell protocols with a greater number of directions).

Conclusion

In this study, we have shown an association between two language functions - picture naming and discourse - and microstructural integrity of the key regions in the language network of patients with subacute post stroke aphasia. Additionally, the integrity of some of these regions before treatment is shown to predict the improvement of linguistic deficits following an experimental treatment. This study paves the road for biomarker driven approach in post stroke aphasia recovery and thus precision language rehabilitation.

Table 2:

Mean diffusivity of left hemisphere regions (mean ± standard deviation)

ID	pSTG	pMTG	pITG	AG	AF	IFOF	ILF	SLF
1	0.782	0.818	0.855	0.794	0.842	0.918	0.903	0.810
2	0.836	0.869	0.891	0.841	0.819	0.869	0.985	0.819
3	0.835	0.860	0.874	0.832	0.816	0.853	0.932	0.819
4	0.761	0.862	0.889	0.791	0.785	0.860	0.792	0.752
5	0.798	0.776	0.903	0.765	0.826	0.841	0.838	0.815
6	0.870	0.865	0.821	0.788	0.857	0.856	0.818	0.812
7	0.700	0.771	0.836	0.632	0.765	0.833	0.773	0.798
8	0.753	0.834	0.805	0.794	0.791	0.873	0.908	0.762
9	0.816	0.857	0.851	0.841	0.828	0.862	0.879	0.800
10	0.751	0.785	0.850	0.798	0.810	0.812	0.827	0.795
11	0.809	0.876	0.944	0.891	0.920	1.021	1.058	0.927
12	0.808	0.834	0.817	0.825	0.905	0.989	0.852	0.879
13	0.808	0.866	0.815	0.926	0.880	0.859	0.909	0.887
14	0.702	0.802	0.825	0.819	0.806	0.857	0.837	0.819
15	0.782	0.849	0.831	0.802	0.903	1.024	1.013	0.914
16	0.779	0.803	0.793	0.753	0.802	0.855	0.887	0.776
17	0.776	0.802	0.843	0.841	0.825	0.861	0.853	0.787
18	0.740	0.818	0.859	0.776	0.856	0.883	0.944	0.806
19	0.785	0.909	0.759	0.826	0.882	0.846	0.874	0.878
20	0.929	1.153	1.160	1.135	0.852	0.865	0.944	0.805
21	0.782	0.807	0.839	0.740	0.884	0.977	0.900	0.876
22	0.890	0.926	0.893	1.068	0.900	0.876	0.885	0.883
	0.795±0.055	0.852±0.079	0.862±0.078	0.831±0.105	0.843±0.044	0.886±0.060	0.891±0.070	0.828±0.049

Data availability statement:

Anonymized data will be made available upon request to the authors, subject to review by the Johns Hopkins University School of Medicine Institutional Review Board resulting in a formal data sharing agreement.