Effects of Transcranial Pulse Stimulation wit ultrasound (TPS) on somatosensory evoked potentials in humans
Eva Matt (Vienna | AT)
Studies using low-intensity transcranial focused ultrasound (tFUS) previously demonstrated effects on somatosensory functions in human subjects such as an attenuation of somatosensory evoked potentials (SEPs) and improved sensory perception . The aim of the current investigation was to test effects of a new MR-navigated focused ultrasound brain stimulation technique – transcranial pulse stimulation (TPS) – on SEPs in healthy participants. Advantages of the new technique are better focusing (no secondary maxima) and lack of brain heating.
10 healthy male subjects were recruited for this randomized, sham-controlled and single-blind study. Using an MR-based navigation system the TPS handpiece was placed with the ultrasound beam focused on the primary somatosensory representation of the right hand (intensity = 0.25 mJ/mm2; frequency = 4 Hz). SEPs were elicited by electrical stimulation of the right median nerve (1.3 Hz, ca. 6.15 min) and recorded at CP3. Data were acquired within one session with a sham and a TPS block (order counterbalanced) with identical runs of alternating SEP recordings (1 baseline run and 3 runs ca. 20 s after TPS/Sham stimulation) and TPS stimulations (3 runs: 10/100/1000 pulses in fixed order). Data preprocessing included filtering, epoching, baseline-correction (-100 to 0 ms) and automated artefact detection (EEGLab ). TPS effects were analyzed with a factorial design with Condition (Sham/Verum) and No. of Pulses (10/100/1000) as factors. Statistical analyses were performed with trial-wise data (all valid epochs) and non-parametrical permutation statistics (1000 permutations, p < 0.025).
The factorial analysis revealed a significant effect of Condition and No. of Pulses, as well as a significant interaction between these factors. The main effect of Condition showed a significantly reduced amplitude for the component N140 (related to conscious perception ) for verum compared to sham that was evident throughout the variable number of pulses. In addition, effects on early components representing sensory processing in primary and secondary somatosensory cortices  were found for the stimulation with 100 (increased N70) and 1000 pulses (increased P27 and N70).
Here, we demonstrated the first sham-controlled evidence for an electrophysiological effect of TPS in humans. TPS effects on EEG components related to cortical somatosensory processing seem to increase with the number of TPS pulses.
fMRI in Patients? Vascular pathophysiological effects exampled on a human translational model
Marco Piccirelli (Zürich | CH)
Task-evoked Blood-oxygenation-level dependent functional MRI (BOLD-fMRI) interpretation shall be improved in Patients showing regions with absent BOLD-fMRI signal activation. Absent BOLD-fMRI signal activation may actually represent false-negative activation due to impaired cerebrovascular reactivity (CVR) of the vascular bed, which has been proven in patients with glioblastoma or stroke.1,2
Nevertheless, CVR relationship to BOLD-fMRI is better studied in subjects with intact neurovascular coupling. Therefore, we artificially controlled the CVR in a healthy population to investigate on a voxel-wise basis the relationship between abnormal CVR and BOLD signals, challenging commonly used linear models.3
Two controlled BOLD-CVR studies on 17 Healthies: 1: at individual resting end-tidal CO2 baseline. 2: +6.0 mmHg CO2 higher. Two BOLD-fMRI finger-tapping experiments were performed at similar normo- and hypercapnic levels.
CVR was determined using an iterative temporal decomposition algorithm.5
Relative BOLD-fMRI signal activation and t-values were calculated for BOLD-CVR and BOLD-fMRI data. For each component of the cerebral motor-network (precentral gyrus, postcentral gyrus, supplementary motor area, cerebellum, fronto-operculum) a voxel-wise quantitative analysis of the impact of BOLD-CVR on BOLD-fMRI was performed.
For the motor-network, the linear correlation coefficient between BOLD-CVR and BOLD-fMRI t-values were significant (p < 0.01) and in the range 0.33-0.55, similar to the correlations between the CVR and fMRI ∆%signal (p < 0.05; range 0.34-0.60).
Interestingly, however, the linear relationship between CVR and fMRI is challenged by our voxel-wise analysis of ∆%signal and t-value change between normo- and hypercapnia. Indeed, the linear relationship holds only for the voxel with supra maximal task activation where the vascularization limits the BOLD signal. For the other voxels, the task-evoked fMRI signal is independent of the CVR, therefore a voxel-wise – as opposed to a global or even ROI-wise – correction/calibration of the BOLD signal is necessary to remove the confound of inhomogeneous CVR over the brain.
We highlight the importance of a complementary BOLD-CVR assessment in addition to task-evoked BOLD fMRI to identify brain areas at risk for false-negative BOLD-fMRI signal activation and to correct for the vascular spatio-temporal filter overlaid on the neuronal activation.
Fast and Direct Estimation of Human Brain Morphometry via Deep Learning
Michael Rebsamen (Bern | CH)
Neurodegenerative and other neurological disorders like Alzheimer’s disease, multiple sclerosis or epilepsy are often associated with structural changes in the brain. Brain morphometry from magnetic resonance imaging (MRI) has proven to be a valuable neuroimaging biomarker for the non-invasive diagnosis and monitoring of such diseases. While manual measurements from MRI are too labor-intensive and error-prone, automated tools often come with a high computational burden (approx. 10h with FreeSurfer), making them hard to use in clinical routine, where time is often an issue.
Recent advances in deep learning for image analysis motivates us to propose a deep learning-based approach for direct estimation (regression) of brain morphometry from MRI. We hypothesize that a neural network can directly predict the volumes of anatomically delineated subcortical regions of interest, and mean thicknesses and curvatures of cortical parcellations. Advantages are the near-time availability of results while maintaining a clinically relevant accuracy.
We propose a deep learning-based approach to directly estimate brain morphometry from T1-weighted MR images. An anonymized dataset of 574 subjects (443 healthy controls and 131 patients with epilepsy) is used for the supervised training of a convolutional neural network (CNN). A "silver-standard" ground truth is generated with FreeSurfer 6.0.
The CNN predicts a total of 165 morphometric measures directly from raw MR images. Analysis of the results using intraclass correlation coefficients showed, in general, good correlation with FreeSurfer generated ground truth data, with some of the regions nearly reaching human inter-rater performance (ICC > 0.75).
Cortical thicknesses predicted by the CNN showed cross-sectional annual age-related gray matter atrophy rates both globally (-0.004 mm/year) and
regionally in agreement with the literature. A population test to dichotomize patients with epilepsy from healthy controls revealed similar effect sizes for structures affecting all subtypes as reported in a large-scale epilepsy study by the ENIGMA consortium (Whelan et al. Brain 2018:141:391).
We have shown the general feasibility of using deep learning to estimate human brain morphometry directly from T1-weighted MRI. A comparison of the results to other publications shows accuracies of comparable magnitudes for the subcortical volumes and cortical thicknesses.
Deep Learning based Detection and Localization of Intracranial Aneurysms in Digital Subtraction Angiography
Manoj Mannil (Zurich | CH)
Background and Purpose: DSA is the gold standard for detecting and characterizing aneurysms. In this study, we assessed the feasibility of a deep learning algorithm for the detection and localization of intracranial aneurysms on DSA images.
Materials and Methods: 706 DSA images were included from a cohort of 240 patients (157 female, mean age 59 years, range 20-92; 83 male patients, mean age 55, range 19-83). 335 (47%) single frame anterior-posterior and lateral images of a DSA series of 187 aneurysms (41 ruptured, 146 unruptured; average size 7 ± 5.3mm, range 1mm to 25mm; total 372 depicted aneurysms) and 371 (53%) aneurysm negative study images were retrospectively analysed regarding the presence and localization of intracranial aneurysms.
The 2D data was split into testing and training sets in a ratio of 4:1 to avoid overfitting with 3D rotational DSA as gold standard. Deep learning was performed using commercial-grade machine learning software (Cognex, ViDi Suite 2.0) based on the open source Tensorflow framework, supervised. Classification results were based solely on unseen test data.
Results: Our deep learning algorithm correctly detected the presence of 292 aneurysms on 335 images (87%) and correctly localized 292 of 372 aneurysms (79%). No size difference was found between detected and undetected aneurysms (6.1 ± 3.9mm vs. 7.3 ± 5.5mm, p=0.65). Intracranial aneurysms were detected and correctly localized with a sensitivity of 79%, a specificity of 79%, a precision of 0.75, a F1 score of 0.77, and an AUC of 0.85.
Conclusion: Deep learning allows for detection and localization of intracranial aneurysms on DSA images.
Hippocampal subfields and Psychiatry
René Seiger (Wien | AT)
The hippocampal formation is regarded a key brain region responsible for memory consolidation and emotional processing. However, recent research also indicates its important role in psychiatric disorders such as major depression . Due to rapid methodological advances in the last years, the delineation of its different nuclei became possible using in-vivo MRI measurements . How the different hippocampal subfields contribute to depression or are affected by different treatment regimens is still not known. Several pipelines have been proposed for the parcellation of the hippocampus, however it is still a matter of debate which approach delivers the most reliable results. In our work, we evaluated different processing strategies for parcellation available within the FreeSurfer software suite, assessed the influence of selective serotonin reuptake inhibitors (SSRIs)  and electroconvulsive therapy (ECT) on hippocampal subfield volumes  and aimed to predict depression using machine learning based on subfield classification .
For the methodological comparison a whole-brain T1-, T2- and a high-resolution T2- weighted sequence at 3T, comprising only the hippocampus, were recorded from 44 healthy subjects at two different time points to assess reliability. The influence of SSRI (escitalopram followed by venlafaxine in case of non-response) treatment was assessed in 20 untreated depressed patients using a T1-weighted sequence and 7T MRI at baseline and after 12 weeks of treatment. The influence of ECT treatment was tested in 14 unipolar treatment resistant depression (TRD) patients after a series of right unilateral ECT treatments using 3T and a T1-weighted sequence. For the machine learning approach 24 depressed patients and 39 healthy controls were measured at 3T (T1-weighted) and “randomForest” within the software “R” was used for prediction. Subfield classification for all investigations above was processed with the hippocampal-subfield pipeline available within FreeSurfer.
Different types of subfield pipelines have already been tested and compared, first results from our methodological evaluation are expected within the next weeks. The influence of SSRI treatment on hippocampal subfields after 12 weeks did not lead to any volumetric changes. Moreover, no alterations after treatment in overall hippocampal volume were found. However, ECT treatment revealed lateralized increases in specific subfields of the right hemisphere, most prominently observed in the granule cell and molecular layer of the dentate gyrus in the hippocampal head and in the hippocampal–amygdaloid transition area (HATA). Interestingly, our machine learning approach also indicated that the HATA seems to play an essential role in the discrimination between depressed patients and healthy controls.
Microstructure of the thalamus in early psychosis and chronic patients based on diffusion weighted magnetic resonance imaging
Yasser Alemán-Gómez (Lausanne | CH)
Introduction: The thalamus has a central role in the pathophysiology of schizophrenia (e.g. deficits in working memory and selective attention). The MedioDorsal (MD), the Pulvinar (Pul) and the anterior nuclei are most frequently reported with reduced volume and number of neurons in post-mortem studies. However how they are affected at the microstructural level is still a matter of debate.
Methods: Twenty-three patients with schizophrenia (SCHZ, 40.18±9.2yo; 18/5 males/females), 35 patients with early psychosis (EP, 24.6±5.68yo; 23/12 males/females) and 62 healthy controls (HC, 31.3±6.57yo; 41/21 males/females) were recruited from the Service of General Psychiatry (Lausanne University Hospital, Switzerland). MRI was performed on a 3-Tesla scanner (Siemens Trio) with T1w and Diffusion Spectrum Imaging (DSI) sequences. We performed an atlas-based segmentation to divide each thalamus in seven clusters in which and Neurite Orientation Dispersion and Density Imaging (NODDI) (based on the DSI) was applied to estimate Orientation Dispersion index (OD), Intracellular Volume Fraction (ICVF) and Isotropic Volume Fraction (ISOVF)). Correlations of these parameters with medication at scanning time were performed. The NODDI parameters were tested in both EP and SCHZ patients for correlations with working memory and speed processing (as assessed by the MATRICS).
Results: Increased OD was found in SCHZ compared to HC. These alterations (increased meanISOVF and meanOD) were located on the right Pul, MD and a cluster containing the medial pulvinar and the centrolateral group. No significant differences were found between EP and HC. Working memory score in EP patients was negatively correlated with ICVF in right MD and survived multiple test correction (r=-0.63, p=.006), whereas correlation between processing speed and ICVF in right Pul was significant but did not survive multiple test correction (r=-0.43, p=.07). By contrast, there were no correlations between NODDI parameters and cognition in SCHZ patients and HC
Discussion: Our results show that the microstructure as assessed by NODDI is significantly affected in MD and Pul of patients with schizophrenia but not in EP, as compared with matched group of control subjects. Interestingly, strong correlations were found between NODDI parameters with cognitive and clinical scores in EP but not in SCHZ. The anatomical location of these anomaiesis in line with post-mortem literature and circuits implicated in schizophrenia.