Limits...
Robust disruptions in electroencephalogram cortical oscillations and large-scale functional networks in autism.

Matlis S, Boric K, Chu CJ, Kramer MA - BMC Neurol (2015)

Bottom Line: The interpretation of prior studies in this area has been limited by mixed results and the lack of validation procedures.Three biomarkers of ASD were identified in the first patient group: (1) reduced posterior/anterior power ratio in the alpha frequency range (8-14 Hz), which we label the "peak alpha ratio", (2) reduced global density in functional networks, and (3) a reduction in the mean connectivity strength of a subset of functional network edges.These results corroborate the general finding that ASD subjects have decreased alpha power gradients and network connectivities compared to control subjects.

View Article: PubMed Central - PubMed

Affiliation: Graduate Program in Neuroscience, Boston University, 677 Beacon st., Boston, MA, 02215, USA. smatlis@bu.edu.

ABSTRACT

Background: Autism spectrum disorders (ASD) are increasingly prevalent and have a significant impact on the lives of patients and their families. Currently, the diagnosis is determined by clinical judgment and no definitive physiological biomarker for ASD exists. Quantitative biomarkers obtainable from clinical neuroimaging data - such as the scalp electroencephalogram (EEG) - would provide an important aid to clinicians in the diagnosis of ASD. The interpretation of prior studies in this area has been limited by mixed results and the lack of validation procedures. Here we use retrospective clinical data from a well-characterized population of children with ASD to evaluate the rhythms and coupling patterns present in the EEG to develop and validate an electrophysiological biomarker of ASD.

Methods: EEG data were acquired from a population of ASD (n = 27) and control (n = 55) children 4-8 years old. Data were divided into training (n = 13 ASD, n = 24 control) and validation (n = 14 ASD, n = 31 control) groups. Evaluation of spectral and functional network properties in the first group of patients motivated three biomarkers that were computed in the second group of age-matched patients for validation.

Results: Three biomarkers of ASD were identified in the first patient group: (1) reduced posterior/anterior power ratio in the alpha frequency range (8-14 Hz), which we label the "peak alpha ratio", (2) reduced global density in functional networks, and (3) a reduction in the mean connectivity strength of a subset of functional network edges. Of these three biomarkers, the first and third were validated in a second group of patients. Using the two validated biomarkers, we were able to classify ASD subjects with 83 % sensitivity and 68 % specificity in a post-hoc analysis.

Conclusions: This study demonstrates that clinical EEG can provide quantitative biomarkers to assist diagnosis of autism. These results corroborate the general finding that ASD subjects have decreased alpha power gradients and network connectivities compared to control subjects. In addition, this study demonstrates the necessity of using statistical techniques to validate EEG biomarkers identified using exploratory methods.

No MeSH data available.


Related in: MedlinePlus

Network analysis reveals that select edges show a significantly diminished density in ASD versus control groups, though not in overall mean density. a Mean density of ASD (blue, Asperger’s in green) and control (red) groups. In the training data, the mean density of the ASD group was significantly lower than the mean density of the control group (p ≤ 0.028). However, in the validation data no significant difference was found (p = 0.50). Error bars represent two standard errors of the mean. b In the training data, no significant difference in degree between ASD and control groups was found at any node location. c The “edge mask”. Edges in the mean control network which were significantly greater than the surrogate control distribution are shown in red (n = 23), while edges in the mean ASD network which were significantly lower than the surrogate ASD distribution are shown in blue (n = 16). Seven edges (shown in orange) were found to distinguish both control from surrogate and ASD from surrogate, and were retrospectively used to form a mask of highly selective edges. d The mask density reveals a significant difference between the ASD group and the control group in training data, as expected (p ≤ 0.0019). The mask density of the ASD group was significantly lower than the mask density of the control group (p ≤ 0.0085) in the validation data as well. e In a retrospective study, the intersection mask density was computed. In both the training and validation data, the ASD intersection mask density was found to be significantly lower than the control intersection mask density (p ≤ 0.0163 in training data, p ≤ 0.0006 in validation data)
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4482270&req=5

Fig3: Network analysis reveals that select edges show a significantly diminished density in ASD versus control groups, though not in overall mean density. a Mean density of ASD (blue, Asperger’s in green) and control (red) groups. In the training data, the mean density of the ASD group was significantly lower than the mean density of the control group (p ≤ 0.028). However, in the validation data no significant difference was found (p = 0.50). Error bars represent two standard errors of the mean. b In the training data, no significant difference in degree between ASD and control groups was found at any node location. c The “edge mask”. Edges in the mean control network which were significantly greater than the surrogate control distribution are shown in red (n = 23), while edges in the mean ASD network which were significantly lower than the surrogate ASD distribution are shown in blue (n = 16). Seven edges (shown in orange) were found to distinguish both control from surrogate and ASD from surrogate, and were retrospectively used to form a mask of highly selective edges. d The mask density reveals a significant difference between the ASD group and the control group in training data, as expected (p ≤ 0.0019). The mask density of the ASD group was significantly lower than the mask density of the control group (p ≤ 0.0085) in the validation data as well. e In a retrospective study, the intersection mask density was computed. In both the training and validation data, the ASD intersection mask density was found to be significantly lower than the control intersection mask density (p ≤ 0.0163 in training data, p ≤ 0.0006 in validation data)

Mentions: After inferring the functional networks from the EEG data (see Methods, Functional network inference and measures), we investigate differences in network topology between the ASD and control groups. Many statistics exist to assess network structure [13, 23]; here we focus on one of the most fundamental – the density – which is computed by summing the number of edges in a network, and then dividing by the number of possible edges. We note that, for the functional networks inferred here, a higher density value indicates an increased level of correlation within the network. The mean density across epochs was calculated for each subject, and averaged within-group (Fig. 3a). In the training group, the ASD population produced a significantly lower mean density than that of the control population (p ≤ 0.028), consistent with some findings in the literature [16, 18, 40, 77–81]. However, in the validation analysis, we found no significant difference in density between the two groups (p = 0.502, Fig. 3a).Fig. 3


Robust disruptions in electroencephalogram cortical oscillations and large-scale functional networks in autism.

Matlis S, Boric K, Chu CJ, Kramer MA - BMC Neurol (2015)

Network analysis reveals that select edges show a significantly diminished density in ASD versus control groups, though not in overall mean density. a Mean density of ASD (blue, Asperger’s in green) and control (red) groups. In the training data, the mean density of the ASD group was significantly lower than the mean density of the control group (p ≤ 0.028). However, in the validation data no significant difference was found (p = 0.50). Error bars represent two standard errors of the mean. b In the training data, no significant difference in degree between ASD and control groups was found at any node location. c The “edge mask”. Edges in the mean control network which were significantly greater than the surrogate control distribution are shown in red (n = 23), while edges in the mean ASD network which were significantly lower than the surrogate ASD distribution are shown in blue (n = 16). Seven edges (shown in orange) were found to distinguish both control from surrogate and ASD from surrogate, and were retrospectively used to form a mask of highly selective edges. d The mask density reveals a significant difference between the ASD group and the control group in training data, as expected (p ≤ 0.0019). The mask density of the ASD group was significantly lower than the mask density of the control group (p ≤ 0.0085) in the validation data as well. e In a retrospective study, the intersection mask density was computed. In both the training and validation data, the ASD intersection mask density was found to be significantly lower than the control intersection mask density (p ≤ 0.0163 in training data, p ≤ 0.0006 in validation data)
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4482270&req=5

Fig3: Network analysis reveals that select edges show a significantly diminished density in ASD versus control groups, though not in overall mean density. a Mean density of ASD (blue, Asperger’s in green) and control (red) groups. In the training data, the mean density of the ASD group was significantly lower than the mean density of the control group (p ≤ 0.028). However, in the validation data no significant difference was found (p = 0.50). Error bars represent two standard errors of the mean. b In the training data, no significant difference in degree between ASD and control groups was found at any node location. c The “edge mask”. Edges in the mean control network which were significantly greater than the surrogate control distribution are shown in red (n = 23), while edges in the mean ASD network which were significantly lower than the surrogate ASD distribution are shown in blue (n = 16). Seven edges (shown in orange) were found to distinguish both control from surrogate and ASD from surrogate, and were retrospectively used to form a mask of highly selective edges. d The mask density reveals a significant difference between the ASD group and the control group in training data, as expected (p ≤ 0.0019). The mask density of the ASD group was significantly lower than the mask density of the control group (p ≤ 0.0085) in the validation data as well. e In a retrospective study, the intersection mask density was computed. In both the training and validation data, the ASD intersection mask density was found to be significantly lower than the control intersection mask density (p ≤ 0.0163 in training data, p ≤ 0.0006 in validation data)
Mentions: After inferring the functional networks from the EEG data (see Methods, Functional network inference and measures), we investigate differences in network topology between the ASD and control groups. Many statistics exist to assess network structure [13, 23]; here we focus on one of the most fundamental – the density – which is computed by summing the number of edges in a network, and then dividing by the number of possible edges. We note that, for the functional networks inferred here, a higher density value indicates an increased level of correlation within the network. The mean density across epochs was calculated for each subject, and averaged within-group (Fig. 3a). In the training group, the ASD population produced a significantly lower mean density than that of the control population (p ≤ 0.028), consistent with some findings in the literature [16, 18, 40, 77–81]. However, in the validation analysis, we found no significant difference in density between the two groups (p = 0.502, Fig. 3a).Fig. 3

Bottom Line: The interpretation of prior studies in this area has been limited by mixed results and the lack of validation procedures.Three biomarkers of ASD were identified in the first patient group: (1) reduced posterior/anterior power ratio in the alpha frequency range (8-14 Hz), which we label the "peak alpha ratio", (2) reduced global density in functional networks, and (3) a reduction in the mean connectivity strength of a subset of functional network edges.These results corroborate the general finding that ASD subjects have decreased alpha power gradients and network connectivities compared to control subjects.

View Article: PubMed Central - PubMed

Affiliation: Graduate Program in Neuroscience, Boston University, 677 Beacon st., Boston, MA, 02215, USA. smatlis@bu.edu.

ABSTRACT

Background: Autism spectrum disorders (ASD) are increasingly prevalent and have a significant impact on the lives of patients and their families. Currently, the diagnosis is determined by clinical judgment and no definitive physiological biomarker for ASD exists. Quantitative biomarkers obtainable from clinical neuroimaging data - such as the scalp electroencephalogram (EEG) - would provide an important aid to clinicians in the diagnosis of ASD. The interpretation of prior studies in this area has been limited by mixed results and the lack of validation procedures. Here we use retrospective clinical data from a well-characterized population of children with ASD to evaluate the rhythms and coupling patterns present in the EEG to develop and validate an electrophysiological biomarker of ASD.

Methods: EEG data were acquired from a population of ASD (n = 27) and control (n = 55) children 4-8 years old. Data were divided into training (n = 13 ASD, n = 24 control) and validation (n = 14 ASD, n = 31 control) groups. Evaluation of spectral and functional network properties in the first group of patients motivated three biomarkers that were computed in the second group of age-matched patients for validation.

Results: Three biomarkers of ASD were identified in the first patient group: (1) reduced posterior/anterior power ratio in the alpha frequency range (8-14 Hz), which we label the "peak alpha ratio", (2) reduced global density in functional networks, and (3) a reduction in the mean connectivity strength of a subset of functional network edges. Of these three biomarkers, the first and third were validated in a second group of patients. Using the two validated biomarkers, we were able to classify ASD subjects with 83 % sensitivity and 68 % specificity in a post-hoc analysis.

Conclusions: This study demonstrates that clinical EEG can provide quantitative biomarkers to assist diagnosis of autism. These results corroborate the general finding that ASD subjects have decreased alpha power gradients and network connectivities compared to control subjects. In addition, this study demonstrates the necessity of using statistical techniques to validate EEG biomarkers identified using exploratory methods.

No MeSH data available.


Related in: MedlinePlus