Deep Learning Achieves Neuroradiologist-Level Performance in Detecting Hydrocephalus Requiring Treatment.

In large clinical centers a small subset of patients present with hydrocephalus that requires surgical treatment. We aimed to develop a screening tool to detect such cases from the head MRI with performance comparable to neuroradiologists. We leveraged 496 clinical MRI exams collected retrospectively at a single clinical site from patients referred for any reason. This diagnostic dataset was enriched to have 259 hydrocephalus cases. A 3D convolutional neural network was trained on 16 manually segmented exams (ten hydrocephalus) and subsequently used to automatically segment the remaining 480 exams and extract volumetric anatomical features. A linear classifier of these features was trained on 240 exams to detect cases of hydrocephalus that required treatment with surgical intervention. Performance was compared to four neuroradiologists on the remaining 240 exams. Performance was also evaluated on a separate screening dataset of 451 exams collected from a routine clinical population to predict the consensus reading from four neuroradiologists using images alone. The pipeline was also tested on an external dataset of 31 exams from a 2nd clinical site. The most discriminant features were the Magnetic Resonance Hydrocephalic Index (MRHI), ventricle volume, and the ratio between ventricle and brain volume. At matching sensitivity, the specificity of the machine and the neuroradiologists did not show significant differences for detection of hydrocephalus on either dataset (proportions test, p > 0.05). ROC performance compared favorably with the state-of-the-art (AUC 0.90-0.96), and replicated in the external validation. Hydrocephalus cases requiring treatment can be detected automatically from MRI in a heterogeneous patient population based on quantitative characterization of brain anatomy with performance comparable to that of neuroradiologists.

Quantitative imaging features predict spinal tap response in normal pressure hydrocephalus.

PURPOSE: Gait improvement following high-volume lumbar puncture (HVLP) and continuous lumbar drain (cLD) is widely used to predict shunt response in patients with suspected normal pressure hydrocephalus (NPH). Here, we investigate differences in MRI volumetric and traditional measures between HVLP/cLD responders and non-responders to identify imaging features that may help predict HVLP/cLD response. METHODS: Eighty-two patients with suspected NPH were studied retrospectively. Gait testing was performed before and immediately/24 h/72 h after HVLP/cLD. A positive response was defined as improvement in gait post-procedure. Thirty-six responders (26 men; mean age 79.3 ± 6.3) and 46 non-responders (25 men; mean age 77.2 ± 6.1) underwent pre-procedure brain MRI including a 3D T1-weighted sequence. Subcortical regional volumes were segmented using FreeSurfer. After normalizing for total intracranial volume, two-way type III ANCOVA test and chi-square test were used to characterize statistical group differences. Evans' index, callosal angle (CA), and disproportionately enlarged subarachnoid space hydrocephalus were assessed. Multivariable logistic regression models were tested using Akaike information criterion to determine which combination of metrics most accurately predicts HVLP/cLD response. RESULTS: Responders and non-responders demonstrated no differences in total ventricular and white/gray matter volumes. CA (men only) and third and fourth ventricular volumes were smaller; and hippocampal volume was larger in responders (p < 0.05). Temporal horns volume correlated with degree of improvement in gait velocity in responders (p = 0.0006). The regression model was 76.8% accurate for HVLP/cLD response. CONCLUSION: CA and third and fourth ventricular volumes and hippocampal volume may serve as potentially useful imaging features that may help predict spinal tap response and hence potentially shunt response.

Surgical debulking of pituitary adenomas improves responsiveness to octreotide lar in the treatment of acromegaly.

BACKGROUND: Studies comparing primary medical treatment of acromegaly with surgery are often non-randomized, and not stratified by illness severity. We prospectively compared primary medical therapy with pituitary surgery in patients with acromegaly. All patients had macroadenomas, at least one random human growth hormone (GH) level ≥12.5 ng/mL, elevated IGF-I levels and failure to suppress GH to <1 ng/mL during an oral glucose tolerance test (oGTT). METHODS: Forty-one patients from seven centers were randomized to primary treatment with octreotide LAR, 30 mg every 4 weeks × 3 months (ARM A, N = 15), or pituitary surgery (ARM B, N = 26) using a 1:2 randomization design. Patients cured by surgery (defined as nadir GH during oGTT <1 ng/mL and normal IGF-I) received no subsequent treatment. Those not cured surgically were then treated with octreotide LAR (SubArm B1) for 3 months. RESULTS: Only one of the 15 patients in ARM A (6.7%) had normalization of both GH and IGF-I. In contrast, 13/26 patients had normalization of both GH and IGF-I after surgery alone (50%). Of the remaining 13 patients who did not normalize with surgery alone, treatment with octreotide LAR resulted in a normal nadir GH and normal serum IGF-I in 7 (53.9%). In total, 20/26 in ARM B (76.9%) experienced normalization of defined biochemical acromegaly parameters. CONCLUSIONS: Pituitary surgery alone was more effective than primary medical treatment (p = 0.006), and the combination of surgery followed by medical therapy was even more effective (p < 0.0001). Subjects treated with medical therapy after surgical debulking had a significant improvement in response rate compared to matched subjects treated with primary medical therapy.

Diagnosis of Normal-Pressure Hydrocephalus: Use of Traditional Measures in the Era of Volumetric MR Imaging.

Purpose To assess the diagnostic performance of the callosal angle (CA) and Evans index (EI) measures and to determine their role versus automated volumetric methods in clinical radiology. Materials and Methods Magnetic resonance (MR) examinations performed before surgery (within 1-5 months of the MR examination) in 36 shunt-responsive patients with normal-pressure hydrocephalus (NPH; mean age, 75 years; age range, 58-87 years; 26 men, 10 women) and MR examinations of age- and sex-matched patients with Alzheimer disease (n = 34) and healthy control volunteers (n = 36) were studied. Three blinded observers independently measured EI and CA for each patient. Volumetric segmentation of global gray matter, white matter, ventricles, and hippocampi was performed by using software. These measures were tested by using multivariable logistic regression models to determine which combination of metrics is most accurate in diagnosis. Results The model that used CA and EI demonstrated 89.6%-93.4% accuracy and average area under the curve of 0.96 in differentiating patients with NPH from patients without NPH (ie, Alzheimer disease and healthy control). The regression model that used volumetric predictors of gray matter and white matter was 94.3% accurate. Conclusion CA and EI may serve as a screening tool to help the radiologist differentiate patients with NPH from patients without NPH, which would allow for designation of patients for further volumetric assessment. © RSNA, 2017.

Motor planning poststroke: impairment in vector-coded reach plans.

Healthy individuals appear to use both vector-coded reach plans that encode movements in terms of their desired direction and extent, and target-coded reach plans that encode the desired endpoint position of the effector. We examined whether these vector and target reach-planning codes are differentially affected after stroke. Participants with stroke and healthy controls made blocks of reaches that were grouped by target location (providing target-specific practice) and by movement vector (providing vector-specific practice). Reach accuracy was impaired in the more affected arm after stroke, but not distinguishable for target- versus vector-grouped reaches. Reach velocity and acceleration were not only impaired in both the less and more affected arms poststroke, but also not distinguishable for target- versus vector-grouped reaches. As previously reported in controls, target-grouped reaches yielded isotropic (circular) error distributions and vector-grouped reaches yielded error distributions elongated in the direction of the reach. In stroke, the pattern of variability was similar. However, the more affected arm showed less elongated error ellipses for vector-grouped reaches compared to the less affected arm, particularly in individuals with right-hemispheric stroke. The results suggest greater impairment to the vector-coded movement-planning system after stroke, and have implications for the development of personalized approaches to poststroke rehabilitation: Motor learning may be enhanced by practice that uses the preserved code or, conversely, by retraining the more impaired code to restore function.

Continuous Versus Sequential Acquisition Head Computed Tomography: A Phantom and Clinical Image Quality Comparative Study.

OBJECTIVE: Sequential computed tomography (CT) technique remains the most common protocol for CT evaluation of the head despite nearly universal adoption of continuous technique for all other body parts. This may be attributable to the belief by radiologists that this technique is superior to helical scanning uniquely for this indication. This study attempts to clarify the issue. METHODS: A phantom evaluation and a prospective randomized clinical image quality trial were performed comparing sequential and helical technique using the most current generation 128-row detector CT scanner. RESULTS: Phantom evaluation demonstrated equivalence between the 2 techniques for low-contrast resolution and line pair spatial discrimination. Continuous scanning provided the highest contrast-to-noise ratio. There was no significant difference between the 2 techniques regarding image quality except for cortical visualization at the cerebral hemispheres, which was subtly but significantly superior for sequential technique. CONCLUSIONS: Head CT image quality for sequential and continuous techniques are virtually equivalent.

Diffusional kurtosis imaging in hydrocephalus.

PURPOSE: Diffusional kurtosis imaging is an advanced diffusion magnetic resonance imaging method that yields, in addition to conventional diffusion information, non-Gaussian diffusion effects, which may allow a more comprehensive characterization of tissue microstructure. The purpose of this study is to use diffusional kurtosis to assess white matter integrity in patients with hydrocephalus and to determine whether changes in kurtosis correlate with the severity of hydrocephalus and leukoaraiosis (LA), a commonly seen comorbidity in hydrocephalus. METHODS: 26 patients with imaging evidence of hydrocephalus and 26 age- and sex- matched subjects with normal ventricular size were retrospectively analyzed. Standard diffusion tensor imaging and diffusional kurtosis metrics were compared between the two groups. Correlation between kurtosis and severity of hydrocephalus and presence and severity of LA was determined. RESULTS: Hydrocephalus patients relative to controls demonstrated statistically significant decrease in all kurtosis metrics in most brain regions studied. The severity of hydrocephalus was associated with greater decrease in kurtosis in the corpus callosum. There was more LA in the hydrocephalus group, and severity of LA was associated with decrease in kurtosis. After controlling for the degree of LA, kurtosis was still decreased in hydrocephalus relative to the controls. CONCLUSION: Diffusional kurtosis imaging detects microstructural changes in the white matter of patients with hydrocephalus. Our results suggest that hydrocephalus plays a role in altering white matter integrity.

Detection of third and sixth cranial nerve palsies with a novel method for eye tracking while watching a short film clip.

OBJECT: Automated eye movement tracking may provide clues to nervous system function at many levels. Spatial calibration of the eye tracking device requires the subject to have relatively intact ocular motility that implies function of cranial nerves (CNs) III (oculomotor), IV (trochlear), and VI (abducent) and their associated nuclei, along with the multiple regions of the brain imparting cognition and volition. The authors have developed a technique for eye tracking that uses temporal rather than spatial calibration, enabling detection of impaired ability to move the pupil relative to normal (neurologically healthy) control volunteers. This work was performed to demonstrate that this technique may detect CN palsies related to brain compression and to provide insight into how the technique may be of value for evaluating neuropathological conditions associated with CN palsy, such as hydrocephalus or acute mass effect. METHODS: The authors recorded subjects' eye movements by using an Eyelink 1000 eye tracker sampling at 500 Hz over 200 seconds while the subject viewed a music video playing inside an aperture on a computer monitor. The aperture moved in a rectangular pattern over a fixed time period. This technique was used to assess ocular motility in 157 neurologically healthy control subjects and 12 patients with either clinical CN III or VI palsy confirmed by neuro-ophthalmological examination, or surgically treatable pathological conditions potentially impacting these nerves. The authors compared the ratio of vertical to horizontal eye movement (height/width defined as aspect ratio) in normal and test subjects. RESULTS: In 157 normal controls, the aspect ratio (height/width) for the left eye had a mean value ± SD of 1.0117 ± 0.0706. For the right eye, the aspect ratio had a mean of 1.0077 ± 0.0679 in these 157 subjects. There was no difference between sexes or ages. A patient with known CN VI palsy had a significantly increased aspect ratio (1.39), whereas 2 patients with known CN III palsy had significantly decreased ratios of 0.19 and 0.06, respectively. Three patients with surgically treatable pathological conditions impacting CN VI, such as infratentorial mass effect or hydrocephalus, had significantly increased ratios (1.84, 1.44, and 1.34, respectively) relative to normal controls, and 6 patients with supratentorial mass effect had significantly decreased ratios (0.27, 0.53, 0.62, 0.45, 0.49, and 0.41, respectively). These alterations in eye tracking all reverted to normal ranges after surgical treatment of underlying pathological conditions in these 9 neurosurgical cases. CONCLUSIONS: This proof of concept series of cases suggests that the use of eye tracking to detect CN palsy while the patient watches television or its equivalent represents a new capacity for this technology. It may provide a new tool for the assessment of multiple CNS functions that can potentially be useful in the assessment of awake patients with elevated intracranial pressure from hydrocephalus or trauma.

Differentiating shunt-responsive normal pressure hydrocephalus from Alzheimer disease and normal aging: pilot study using automated MRI brain tissue segmentation.

Evidence suggests that normal pressure hydrocephalus (NPH) is underdiagnosed in day to day radiologic practice, and differentiating NPH from cerebral atrophy due to other neurodegenerative diseases and normal aging remains a challenge. To better characterize NPH, we test the hypothesis that a prediction model based on automated MRI brain tissue segmentation can help differentiate shunt-responsive NPH patients from cerebral atrophy due to Alzheimer disease (AD) and normal aging. Brain segmentation into gray and white matter (GM, WM), and intracranial cerebrospinal fluid was derived from pre-shunt T1-weighted MRI of 15 shunt-responsive NPH patients (9 men, 72.6 ± 8.0 years-old), 17 AD patients (10 men, 72.1 ± 11.0 years-old) chosen as a representative of cerebral atrophy in this age group; and 18 matched healthy elderly controls (HC, 7 men, 69.7 ± 7.0 years old). A multinomial prediction model was generated based on brain tissue volume distributions. GM decrease of 33% relative to HC characterized AD (P < 0.005). High preoperative ventricular and near normal GM volumes characterized NPH. A multinomial regression model based on gender, GM and ventricular volume had 96.3% accuracy differentiating NPH from AD and HC. In conclusion, automated MRI brain tissue segmentation differentiates shunt-responsive NPH with high accuracy from atrophy due to AD and normal aging. This method may improve diagnosis of NPH and improve our ability to distinguish normal from pathologic aging.

In vivo 7 Tesla imaging of the dentate granule cell layer in schizophrenia.

PURPOSE: The hippocampus is central to the pathophysiology of schizophrenia. Histology shows abnormalities in the dentate granule cell layer (DGCL), but its small size (~100 µm thickness) has precluded in vivo human studies. We used ultra high field magnetic resonance imaging (MRI) to compare DGCL morphology of schizophrenic patients to matched controls. METHOD: Bilateral hippocampi of 16 schizophrenia patients (10 male) 40.7 ± 10.6 years old (mean ± standard deviation) were imaged at 7 Tesla MRI with heavily T2*-weighted gradient-echo sequence at 232 µm in-plane resolution (0.08 µL image voxels). Fifteen matched controls (8 male, 35.6 ± 9.4 years old) and one ex vivo post mortem hippocampus (that also underwent histopathology) were scanned with same protocol. Three blinded neuroradiologists rated each DGCL on a qualitative scale of 1 to 6 (from "not discernible" to "easily visible, appearing dark gray or black") and mean left and right DGCL scores were compared using a non-parametric Mann-Whitney test. RESULTS: MRI identification of the DGCL was validated with histopathology. Mean right and left DGCL ratings in patients (3.2 ± 1.0 and 3.5 ± 1.2) were not statistically different from those of controls (3.9 ± 1.1 and 3.8 ± 0.8), but patients had a trend for lower right DGCL score (p = 0.07), which was significantly associated with patient diagnosis (p = 0.05). The optimal 48% sensitivity and 80% specificity for schizophrenia were achieved with a DGCL rating of ≤2. CONCLUSION: Decreased contrast in the right DGCL in schizophrenia was predictive of schizophrenia diagnosis. Better utility of this metric as a schizophrenia biomarker may be achieved in future studies of patients with homogeneous disease subtypes and progression rates.