Functional connectivity model identifies MS impairment

WEST PALM BEACH, FLA. – A machine learning model that combines data on the brain’s functional connectivity with clinical information such as age, sex and disease duration shows the potential to provide an accurate assessment of clinical impairment in patients with multiple sclerosis (MS).

“This is the first study to show that dynamic functional connectivity is useful to identify the impairment level in MS, and can be used for personalized treatment by clinicians,” first author Ceren Tozlu, PhD, of Weill Cornell Medicine, New York, said in an interview.

“We found out that structural connectivity is the most important feature that distinguishes MS patients from healthy controls, while dynamic functional connectivity was more discriminative compared to the static functional connectivity in MS patient classification regarding their impairment level.”

The findings were presented at the meeting held by the Americas Committee for Treatment and Research in Multiple Sclerosis.

Statistical assessment of the clinical impairment of MS using MRI is hindered by a relatively weak correlation between the impairment and disease burden, such as lesion load.

However, the brain’s functional connectivity network, which is indicative of the disruption of the transmission of signals of gray matter regions, could provide a deeper understanding of connectome-level mechanisms that underlie variability in MS-related impairments, Dr. Tozlu and colleagues say.

With no previous study pulling together multimodal imaging data including static and dynamic functional connectivity to classify MS patients with a clinically significant impairment versus non–clinically significant impairment, Dr. Tozlu and the team sought to build a machine-learning–based model to do so.

For the study, they enrolled 79 patients with MS, including 42 with Expanded Disability Status scores of 2 or higher, representing clinically significant impairment at baseline.

The patients, who had a mean age of 45 years, were 66% female and had a mean disease duration of 12.48 years. The ensemble model that was used incorporated functional connectivity and a clinical dataset of age, sex, and disease duration. Functional connectivity was measured by evaluating blood oxygen level dependent (BOLD) signal activity between 86 FreeSurfer-based gray matter regions.

“Functional connectivity is a statistical correlation (Pearson’s correlation coefficient) between two time series of BOLD signals measured on two distinct region of interest of the brain during MRI scan,” Dr. Tozlu explained. “In our study, BOLD time series were measured using resting-state functional MRI technique that last 7 minutes.”

The ensemble model was able to classify low-adapting MS patients with an area under ROC curve (AUC) of 0.638 and a balanced accuracy of 0.659. The model performed well in accurately classifying the MS patients with clinically significant impairment with a sensitivity of 0.719.

“The models in which we applied functional and structural connectivity showed a high performance in classifying MS patients regarding their impairment level,” Dr. Tozlu said.

She noted that “these models may be extended to predict change in impairment level in a longitudinal study, for instance, identifying MS patients who may have a clinically significant impairment.”

In further evaluating which particular functional connections were most related to MS disease activity, Dr. Tozlu and colleagues found the most discriminative areas were between the right superior parietal and right inferior temporal, between right lateral occipital and left pericalcarine, and between right pericalcarine and right side of frontal pole.

If further validated, the approach could have important, broader clinical implications, Dr. Tozlu said.

“If the validation of these models on a larger dataset is successful, this model may be used to decide for personalized treatment,” Dr. Tozlu added. “The model could offer guidance in providing more powerful treatment for MS patients who may have a clinically significant impairment and less powerful treatment for MS patients who may not have a clinically significant impairment in order to avoid the side effects of treatments.

“Therefore, we believe that dynamics in functional connectivity should be taken into account in the next studies in MS.”

In commenting on the research, Eric Klawiter, MD, associate professor of neurology, Harvard Medical School and associate neurologist at Massachusetts General Hospital, both in Boston, said the findings offer valuable insights in the use of machine learning and MS imaging.

“This research shows very nicely the power of machine learning and connectivity techniques to differentiate MS phenotypes based on disability level,” he said in an interview.

“The future direction of this work is to develop predictive markers for disability progression and this would have significant impact in how we evaluate newly diagnosed patients and counsel their treatment decisions.”

Dr. Tozlu and Dr. Klawiter had no disclosures to report.

SOURCE: Tozlu C et al. ACTRIMS Forum 2020. Abstract P025.

WEST PALM BEACH, FLA. – A machine learning model that combines data on the brain’s functional connectivity with clinical information such as age, sex and disease duration shows the potential to provide an accurate assessment of clinical impairment in patients with multiple sclerosis (MS).

“This is the first study to show that dynamic functional connectivity is useful to identify the impairment level in MS, and can be used for personalized treatment by clinicians,” first author Ceren Tozlu, PhD, of Weill Cornell Medicine, New York, said in an interview.

“We found out that structural connectivity is the most important feature that distinguishes MS patients from healthy controls, while dynamic functional connectivity was more discriminative compared to the static functional connectivity in MS patient classification regarding their impairment level.”

The findings were presented at the meeting held by the Americas Committee for Treatment and Research in Multiple Sclerosis.

Statistical assessment of the clinical impairment of MS using MRI is hindered by a relatively weak correlation between the impairment and disease burden, such as lesion load.

However, the brain’s functional connectivity network, which is indicative of the disruption of the transmission of signals of gray matter regions, could provide a deeper understanding of connectome-level mechanisms that underlie variability in MS-related impairments, Dr. Tozlu and colleagues say.

With no previous study pulling together multimodal imaging data including static and dynamic functional connectivity to classify MS patients with a clinically significant impairment versus non–clinically significant impairment, Dr. Tozlu and the team sought to build a machine-learning–based model to do so.

For the study, they enrolled 79 patients with MS, including 42 with Expanded Disability Status scores of 2 or higher, representing clinically significant impairment at baseline.

The patients, who had a mean age of 45 years, were 66% female and had a mean disease duration of 12.48 years. The ensemble model that was used incorporated functional connectivity and a clinical dataset of age, sex, and disease duration. Functional connectivity was measured by evaluating blood oxygen level dependent (BOLD) signal activity between 86 FreeSurfer-based gray matter regions.

“Functional connectivity is a statistical correlation (Pearson’s correlation coefficient) between two time series of BOLD signals measured on two distinct region of interest of the brain during MRI scan,” Dr. Tozlu explained. “In our study, BOLD time series were measured using resting-state functional MRI technique that last 7 minutes.”

The ensemble model was able to classify low-adapting MS patients with an area under ROC curve (AUC) of 0.638 and a balanced accuracy of 0.659. The model performed well in accurately classifying the MS patients with clinically significant impairment with a sensitivity of 0.719.

“The models in which we applied functional and structural connectivity showed a high performance in classifying MS patients regarding their impairment level,” Dr. Tozlu said.

She noted that “these models may be extended to predict change in impairment level in a longitudinal study, for instance, identifying MS patients who may have a clinically significant impairment.”

In further evaluating which particular functional connections were most related to MS disease activity, Dr. Tozlu and colleagues found the most discriminative areas were between the right superior parietal and right inferior temporal, between right lateral occipital and left pericalcarine, and between right pericalcarine and right side of frontal pole.

If further validated, the approach could have important, broader clinical implications, Dr. Tozlu said.

“If the validation of these models on a larger dataset is successful, this model may be used to decide for personalized treatment,” Dr. Tozlu added. “The model could offer guidance in providing more powerful treatment for MS patients who may have a clinically significant impairment and less powerful treatment for MS patients who may not have a clinically significant impairment in order to avoid the side effects of treatments.

“Therefore, we believe that dynamics in functional connectivity should be taken into account in the next studies in MS.”

In commenting on the research, Eric Klawiter, MD, associate professor of neurology, Harvard Medical School and associate neurologist at Massachusetts General Hospital, both in Boston, said the findings offer valuable insights in the use of machine learning and MS imaging.

“This research shows very nicely the power of machine learning and connectivity techniques to differentiate MS phenotypes based on disability level,” he said in an interview.

“The future direction of this work is to develop predictive markers for disability progression and this would have significant impact in how we evaluate newly diagnosed patients and counsel their treatment decisions.”

Dr. Tozlu and Dr. Klawiter had no disclosures to report.

SOURCE: Tozlu C et al. ACTRIMS Forum 2020. Abstract P025.

WEST PALM BEACH, FLA. – A machine learning model that combines data on the brain’s functional connectivity with clinical information such as age, sex and disease duration shows the potential to provide an accurate assessment of clinical impairment in patients with multiple sclerosis (MS).

“This is the first study to show that dynamic functional connectivity is useful to identify the impairment level in MS, and can be used for personalized treatment by clinicians,” first author Ceren Tozlu, PhD, of Weill Cornell Medicine, New York, said in an interview.

“We found out that structural connectivity is the most important feature that distinguishes MS patients from healthy controls, while dynamic functional connectivity was more discriminative compared to the static functional connectivity in MS patient classification regarding their impairment level.”

The findings were presented at the meeting held by the Americas Committee for Treatment and Research in Multiple Sclerosis.

Statistical assessment of the clinical impairment of MS using MRI is hindered by a relatively weak correlation between the impairment and disease burden, such as lesion load.

However, the brain’s functional connectivity network, which is indicative of the disruption of the transmission of signals of gray matter regions, could provide a deeper understanding of connectome-level mechanisms that underlie variability in MS-related impairments, Dr. Tozlu and colleagues say.

With no previous study pulling together multimodal imaging data including static and dynamic functional connectivity to classify MS patients with a clinically significant impairment versus non–clinically significant impairment, Dr. Tozlu and the team sought to build a machine-learning–based model to do so.

For the study, they enrolled 79 patients with MS, including 42 with Expanded Disability Status scores of 2 or higher, representing clinically significant impairment at baseline.

The patients, who had a mean age of 45 years, were 66% female and had a mean disease duration of 12.48 years. The ensemble model that was used incorporated functional connectivity and a clinical dataset of age, sex, and disease duration. Functional connectivity was measured by evaluating blood oxygen level dependent (BOLD) signal activity between 86 FreeSurfer-based gray matter regions.

“Functional connectivity is a statistical correlation (Pearson’s correlation coefficient) between two time series of BOLD signals measured on two distinct region of interest of the brain during MRI scan,” Dr. Tozlu explained. “In our study, BOLD time series were measured using resting-state functional MRI technique that last 7 minutes.”

The ensemble model was able to classify low-adapting MS patients with an area under ROC curve (AUC) of 0.638 and a balanced accuracy of 0.659. The model performed well in accurately classifying the MS patients with clinically significant impairment with a sensitivity of 0.719.

“The models in which we applied functional and structural connectivity showed a high performance in classifying MS patients regarding their impairment level,” Dr. Tozlu said.

She noted that “these models may be extended to predict change in impairment level in a longitudinal study, for instance, identifying MS patients who may have a clinically significant impairment.”

In further evaluating which particular functional connections were most related to MS disease activity, Dr. Tozlu and colleagues found the most discriminative areas were between the right superior parietal and right inferior temporal, between right lateral occipital and left pericalcarine, and between right pericalcarine and right side of frontal pole.

If further validated, the approach could have important, broader clinical implications, Dr. Tozlu said.

“If the validation of these models on a larger dataset is successful, this model may be used to decide for personalized treatment,” Dr. Tozlu added. “The model could offer guidance in providing more powerful treatment for MS patients who may have a clinically significant impairment and less powerful treatment for MS patients who may not have a clinically significant impairment in order to avoid the side effects of treatments.

“Therefore, we believe that dynamics in functional connectivity should be taken into account in the next studies in MS.”

In commenting on the research, Eric Klawiter, MD, associate professor of neurology, Harvard Medical School and associate neurologist at Massachusetts General Hospital, both in Boston, said the findings offer valuable insights in the use of machine learning and MS imaging.

“This research shows very nicely the power of machine learning and connectivity techniques to differentiate MS phenotypes based on disability level,” he said in an interview.

“The future direction of this work is to develop predictive markers for disability progression and this would have significant impact in how we evaluate newly diagnosed patients and counsel their treatment decisions.”

Dr. Tozlu and Dr. Klawiter had no disclosures to report.

SOURCE: Tozlu C et al. ACTRIMS Forum 2020. Abstract P025.