Eye Tracking for Deep Learning Segmentation Using Convolutional Neural Networks.

Deep learning with convolutional neural networks (CNNs) has experienced tremendous growth in multiple healthcare applications and has been shown to have high accuracy in semantic segmentation of medical (e.g., radiology and pathology) images. However, a key barrier in the required training of CNNs is obtaining large-scale and precisely annotated imaging data. We sought to address the lack of annotated data with eye tracking technology. As a proof of principle, our hypothesis was that segmentation masks generated with the help of eye tracking (ET) would be very similar to those rendered by hand annotation (HA). Additionally, our goal was to show that a CNN trained on ET masks would be equivalent to one trained on HA masks, the latter being the current standard approach. Step 1: Screen captures of 19 publicly available radiologic images of assorted structures within various modalities were analyzed. ET and HA masks for all regions of interest (ROIs) were generated from these image datasets. Step 2: Utilizing a similar approach, ET and HA masks for 356 publicly available T1-weighted postcontrast meningioma images were generated. Three hundred six of these image + mask pairs were used to train a CNN with U-net-based architecture. The remaining 50 images were used as the independent test set. Step 1: ET and HA masks for the nonneurological images had an average Dice similarity coefficient (DSC) of 0.86 between each other. Step 2: Meningioma ET and HA masks had an average DSC of 0.85 between each other. After separate training using both approaches, the ET approach performed virtually identically to HA on the test set of 50 images. The former had an area under the curve (AUC) of 0.88, while the latter had AUC of 0.87. ET and HA predictions had trimmed mean DSCs compared to the original HA maps of 0.73 and 0.74, respectively. These trimmed DSCs between ET and HA were found to be statistically equivalent with a p value of 0.015. We have demonstrated that ET can create segmentation masks suitable for deep learning semantic segmentation. Future work will integrate ET to produce masks in a faster, more natural manner that distracts less from typical radiology clinical workflow.

High-pitch dual-source CT angiography of supra-aortic arteries: assessment of image quality and radiation dose.

INTRODUCTION: High-pitch CT angiography (CTA) is a recent innovation that allows significant shortening of scan time with volume coverage of 43 mm per second. The aim of our study was to assess this technique in CTA of the head and neck. METHODS: CTA of supra-aortic arteries was performed in 50 patients using two acquisition protocols: conventional single-source 64-slice (pitch 1.2) and high-pitch dual-source 128-slice CT (pitch 3.2). Subjective and objective image quality of supra-aortic vessel ostia as well as intra- and extra-cranial segments was retrospectively assessed by blinded readers and radiation dose compared between the two protocols. RESULTS: Conventional and high-pitch CTA achieved comparable signal-to-noise ratios in arterial (54.3 ± 16.5 versus 57.3 ± 14.8; p = 0.50) and venous segments (15.8 ± 6.7 versus 18.9 ± 8.9; p = 0.21). High-pitch scanning was, however, associated with sharper delineation of vessel contours and image quality significantly improved at the level of supra-aortic vessel ostia (p < 0.0001) as well as along the brachiocephalic trunk (p < 0.0001), the subclavian arteries (p < 0.0001), proximal common carotid arteries (p = 0.01), and vertebral V1 segments (p < 0.0001). Using the high-pitch mode, the dose-length product was reduced by about 35% (218.2 ± 30 versus 141.8 ± 20 mGy × cm). CONCLUSIONS: Due to elimination of transmitted cardiac motion, high-pitch CTA of the neck improves image quality in the proximity of the aortic arch while significantly lowering radiation dose. The technique thus qualifies as a promising alternative to conventional spiral CTA and may be particularly useful for identification of ostial stenosis.

The Cochlea in Branchio-Oto-Renal Syndrome: An Objective Method for the Diagnosis of Offset Cochlear Turns.

BACKGROUND AND PURPOSE: An "unwound" or "offset" cochlea has been described as a characteristic imaging feature in patients with branchio-oto-renal syndrome, and recently recognized to be associated in particular to those with EYA1 gene mutations. Determination of this feature has traditionally relied on subjective visual assessment. Our aim was to establish an objective assessment method for cochlear offset (the cochlear turn alignment ratio) and determine an optimal cutoff turn alignment ratio value that separates individuals with EYA1-branchio-oto-renal syndrome from those with SIX1-branchio-oto-renal syndrome and healthy controls. MATERIALS AND METHODS: Temporal bone CT or MR imaging from 40 individuals with branchio-oto-renal syndrome and 40 controls was retrospectively reviewed. Cochlear offset was determined visually by 2 independent blinded readers and then quantitatively via a standardized technique yielding the cochlear turn alignment ratio. The turn alignment ratio values were compared between cochleae qualitatively assessed as "not offset" and "offset." Receiver operating characteristic analysis was used to determine the ability of the turn alignment ratio to differentiate between these populations and an optimal cutoff turn alignment ratio value. Cochlear offset and turn alignment ratio values were analyzed for each branchio-oto-renal syndrome genotype subpopulation and for controls. RESULTS: The turn alignment ratio can accurately differentiate between cochleae with and without an offset (P < .001). The optimal cutoff value separating these populations was 0.476 (sensitivity = 1, specificity = 0.986, J = 0.986). All except 1 cochlea among the EYA1-branchio-oto-renal syndrome subset and all with unknown genotype branchio-oto-renal syndrome had a cochlear offset and a turn alignment ratio of <0.476. All except 1 cochlea among the SIX1-branchio-oto-renal syndrome subset and all controls had no offset and a turn alignment ratio of >0.476. CONCLUSIONS: There is a statistically significant difference in turn alignment ratios between offset and nonoffset cochleae, with an optimal cutoff of 0.476. This cutoff value allows excellent separation of EYA1-branchio-oto-renal syndrome from SIX1-branchio-oto-renal syndrome and from individuals without branchio-oto-renal syndrome or sensorineural hearing loss. The turn alignment ratio is a reliable and objective metric that can aid in the imaging evaluation of branchio-oto-renal syndrome.

Re-Examining the Cochlea in Branchio-Oto-Renal Syndrome: Genotype-Phenotype Correlation.

BACKGROUND AND PURPOSE: Temporal bone imaging plays an important role in the work-up of branchio-oto-renal syndrome. Previous reports have suggested that the unwound or offset cochlea is a highly characteristic marker for branchio-oto-renal syndrome. Our goals were to examine the prevalence of this finding in a branchio-oto-renal syndrome cohort and analyze genetic-phenotypic associations not previously established. MATERIALS AND METHODS: This multicenter retrospective study included 38 ears in 19 unrelated individuals with clinically diagnosed branchio-oto-renal syndrome and confirmed mutations in the EYA1 or SIX1 genes. Two blinded neuroradiologists independently reviewed and documented temporal bone imaging findings in 13 categories for each ear. Imaging phenotypes were correlated with genotypes. RESULTS: There was excellent interrater agreement for all 13 phenotypic categories (κ ≥ 0.80). Of these, 9 categories showed statistically significant differences between patients with EYA1-branchio-oto-renal syndrome and SIX1-branchio-oto-renal syndrome. Cochlear offset was present in 100% of patients with EYA1-branchio-oto-renal syndrome, but in only 1 ear (12.5%) among patients with SIX1-branchio-oto-renal syndrome. A short thorny appearance of the cochlear apical turn was observed in most patients with SIX1-branchio-oto-renal syndrome. CONCLUSIONS: An offset cochlea is associated with the EYA1-branchio-oto-renal syndrome genotype. The SIX1-branchio-oto-renal syndrome genotype is associated with a different cochlear phenotype that almost always is without offset and has a short thorny tip as the apical turn. Therefore, cochlear offset is not a characteristic marker for all patients with branchio-oto-renal syndrome. The lack of a cochlear offset in a patient with clinically suspected branchio-oto-renal syndrome does not exclude the diagnosis and, in fact, may be predictive of the SIX1 genotype.

Protuberant Fibro-Osseous Lesion of the Temporal Bone: "Bullough Bump"-Multimodality Imaging Case Series and Literature Review.

A handful of cases of protuberant fibro-osseous lesions of the temporal bones have been described in the literature to date, with primary focus on the pathologic features. Here we review 3 cases of pathology-proved protuberant fibro-osseous lesions of the temporal bone and include a literature review with a focus on the imaging features. While rare, these lesions have near-pathognomonic imaging features defined by a location at the cortex of the outer table of the temporal bone at the occipitomastoid suture, lack of involvement of the underlying marrow, variable mineralization, and MR signal characteristics atypical of a chondroid lesion. One case in this series was FDG-avid and had occasional mitotic features, possibly reflecting an aggressive variant. Neuroradiologists should be familiar with this benign diagnosis to aid in timely identification and avoid unnecessary additional imaging.

From the Eye of the Storm: Multi-Institutional Practical Perspectives on Neuroradiology from the COVID-19 Outbreak in New York City.

During the Severe Acute Respiratory Syndrome Coronavirus 2 (COVID-19) pandemic, neuroradiology practices have experienced a paradigm shift in practice, which affected everything from staffing, workflow, work volumes, conferences, resident and fellowship education, and research. This article highlights adaptive strategies that were undertaken at the epicenter of the outbreak in New York City during the past 4-6 weeks, as experienced by 5 large neuroradiology academic departments.

Imaging characteristics of recurrent pleomorphic adenoma of the parotid gland.

BACKGROUND AND PURPOSE: Pleomorphic adenoma is the most common parotid neoplasm. It has a relatively high rate of recurrence after surgery. Imaging features of recurrent parotid pleomorphic adenoma have been infrequently reported in the radiology literature. In the present study, we reviewed our institutional experience with imaging of recurrent parotid pleomorphic adenomas. MATERIALS AND METHODS: Retrospective imaging and chart review of 24 patients (9 men, 15 women; age, 29-63 years) with recurrent pleomorphic adenoma of the parotid were performed. Images were reviewed as to the margins, site, multiplicity, signal intensity, and enhancement pattern of the recurrent lesions by 2 neuroradiologists. RESULTS: We found a high incidence of multiple lesions in the operative bed, most of which were extremely bright on T2-weighted images; some of the lesions demonstrated a cystic appearance with peripheral enhancement. We also noted tiny lesions in the subcutaneous fat and in regions distant from the immediate operative bed. CONCLUSION: Although recurrent disease is usually clinically apparent, sometimes prior surgical history is lacking or recurrence may be detected on routine surveillance imaging after initial surgery. The presence of solitary or multiple subcutaneous T2 hyperintense lesions in the operative bed, subcutaneous fat, and/or spaces adjacent to the parotid in patients with prior parotidectomy for pleomorphic adenoma is consistent with the diagnosis of neoplastic recurrence. The radiologist should maintain a high index of suspicion for recurrent pleomorphic adenoma with this characteristic imaging appearance.

Spectrum of Third Window Abnormalities: Semicircular Canal Dehiscence and Beyond.

Third window abnormalities are defects in the integrity of the bony structure of the inner ear, classically producing sound-/pressure-induced vertigo (Tullio and Hennebert signs) and/or a low-frequency air-bone gap by audiometry. Specific anatomic defects include semicircular canal dehiscence, perilabyrinthine fistula, enlarged vestibular aqueduct, dehiscence of the scala vestibuli side of the cochlea, X-linked stapes gusher, and bone dyscrasias. We discuss these various entities and provide key examples from our institutional teaching file with a discussion of symptomatology, temporal bone CT, audiometry, and vestibular-evoked myogenic potentials.

"Finding a Voice": Imaging Features after Phonosurgical Procedures for Vocal Fold Paralysis.

Altered communication (hoarseness, dysphonia, and breathy voice) that can result from vocal fold paralysis, secondary to numerous etiologies, may be amenable to surgical restoration. In this article, both traditional and cutting-edge phonosurgical procedures targeting the symptoms resulting from vocal fold paralysis are reviewed, with emphasis on the characteristic imaging appearances of various injectable materials, implants, and augmentation procedures used in the treatment of vocal fold paralysis. In addition, complications of injection laryngoplasty and medialization laryngoplasty are illustrated. Familiarity with the expected imaging changes following treatment of vocal fold paralysis may prevent the misinterpretation of posttreatment changes as pathology. Identifying common complications related to injection laryngoplasty and localization of displaced implants is crucial in determining specific management in patients who have undergone phonosurgical procedures for the management of vocal fold paralysis.

Computing diffusion rates in T2-dark hematomas and areas of low T2 signal.

BACKGROUND AND PURPOSE: It has been suggested that restricted diffusion is present within hematomas with intact red cell membranes; however, computing apparent diffusion coefficient (ADC) values in areas of low T2 signal can be problematic. Our purpose was to show the pitfalls of measuring diffusion within hematomas with intracellular blood products and to present a framework based on the properties of expected values for computing ADC values from regions with signal intensities close to that of the background noise (ie, T2-dark hematomas). METHODS: Twelve patients with intracranial hematomas who had undergone diffusion imaging were retrospectively identified during a 2-year period (four intracellular oxyhemoglobin, seven intracellular deoxyhemoglobin, one intracellular methemoglobin). Regions of interest were drawn on the hematomas, the contralateral white matter, and over the background. ADC values were computed using a variety of methods: 1) using expected values incorporating the variance of the background, 2) computing the mean of the regions of interest before taking the natural log, 3) masking negative values, and 4) masking the background at 0.5% increments from 0.5 to 5.5% and including the masked voxels (an intrinsically flawed method). Two-tailed Student's t test was performed between the white matter and the hematoma ADC values. RESULTS: There was no statistically significant difference between the hematomas and the white matter for methods 1 through 3 (P = .14, P = .23, and P = .83, respectively). Only method 4 revealed a statistically significant difference, beginning at 0.5% masking (P = .04) and becoming progressively more significant with increased masking (P = 4.14 x 10(-7) at 5.5% masking). The effect of masking was limited to the T2-dark hematomas. CONCLUSION: There is no restriction of diffusion for in vivo hematomas with intracellular blood products. The T2 blackout effect for T2-dark hematomas on diffusion-weighted images should not be interpreted as fast diffusion. The method of expected values can be used to obtain measurements for regions with signal intensities near the background noise. Using literature values for RBC self-diffusion, we computed lower limits of diffusion for hematomas with intracellular blood products to be 0.3 x 10(-3) mm2/s.

MRI findings of isolated beta-2 microglobulin amyloidosis presenting as a cervical spine mass. Case report and review of literature.

A patient without prior history of renal failure or hemodialysis presented with weakness of the extremities. On magnetic resonance imaging (MRI), a peri-odontoid soft-tissue mass hypointense to muscle on both T1- and T2-weighted images was noted without obvious bony destruction. Pathological examination revealed degenerated fibrocartilage which stained for amyloid. Further, peroxidase staining for beta-2 microglobulin was positive. We describe the first case of a beta-2 microglobulin peri-odontoid amyloidoma without preceding history of hemodialysis and a review of literature is performed.

Imaging findings in auto-atticotomy.

BACKGROUND AND PURPOSE: An acquired attic cholesteatoma may spontaneously drain externally into the external auditory canal, leaving a cavity in the attic with the shape of the original cholesteatoma but now filled with air, a phenomenon referred to as "nature's atticotomy" or auto-atticotomy. We describe and quantify the CT appearance of the auto-atticotomy cavity as it pertains to the appearance of the scutum and the lateral attic wall. MATERIALS AND METHODS: Twenty-one patients with erosion of the scutum and loss of the lower attic wall on MDCT were identified during a 5-year span. Images were assessed for measureable widening of the space between the ossicles and the lower lateral attic wall in the axial and coronal planes. Three measurements of the lateral attic were made on the axial images. Findings were compared with the same measurements in 20 control subjects. RESULTS: The 21 patients had a characteristic blunting of the scutum with loss of the lower lateral attic wall and widening of the lateral attic, consistent with an auto-atticotomy. There was a statistically significant (P < .001) widening of the lateral attic dimensions in the axial plane in the patients with auto-atticotomy. CONCLUSIONS: Spontaneously evacuated cholesteatoma may mimic a surgical atticotomy on MDCT. Scutal erosion and attic enlargement with a smoothly contoured bony remodeling of the lower lateral attic wall in a patient with no history of surgery suggest that a cholesteatoma was previously present and spontaneously drained.

Imaging appearance of the lateral rectus-superior rectus band in 100 consecutive patients without strabismus.

BACKGROUND AND PURPOSE: The lateral rectus-superior rectus band is an orbital connective tissue structure that has been implicated in a form of strabismus termed sagging eye syndrome. Our purpose was to define the normal MR imaging and CT appearance of this band in patients without strabismus. MATERIALS AND METHODS: Orbital MR imaging and CT examinations in 100 consecutive patients without strabismus were evaluated. Readers graded the visibility of the lateral rectus-superior rectus band on coronal T1WI, coronal STIR, and coronal CT images. Readers determined whether the band demonstrated superotemporal bowing or any discontinuities and whether a distinct lateral levator aponeurosis was seen. Reader agreement was assessed by κ coefficients. Association between imaging metrics and patient age/sex was calculated by using the Fisher exact test. RESULTS: The lateral rectus-superior rectus band was visible in 95% of coronal T1WI, 68% of coronal STIR sequences, and 70% of coronal CT scans. Ninety-five percent of these bands were seen as a continuous, arc-like structure extending from the superior rectus/levator palpebrae muscle complex to the lateral rectus muscle; 24% demonstrated superotemporal bowing; and in 82% of orbits, a distinct lateral levator aponeurosis was visible. Increasing patient age was negatively associated with lateral rectus-superior rectus band visibility (P=.03), positively associated with lateral rectus-superior rectus band superotemporal bowing (P=.03), and positively associated with lateral levator aponeurosis visibility (P=.01). CONCLUSIONS: The lateral rectus-superior rectus band is visible in most patients without strabismus on coronal T1WI. The age effect with respect to its visibility and superotemporal bowing could represent age-related connective tissue degeneration.

Syndromes of the first and second branchial arches, part 2: syndromes.

A variety of congenital syndromes affecting the face occur due to defects involving the first and second BAs. Radiographic evaluation of craniofacial deformities is necessary to define aberrant anatomy, plan surgical procedures, and evaluate the effects of craniofacial growth and surgical reconstructions. High-resolution CT has proved vital in determining the nature and extent of these syndromes. The radiologic evaluation of syndromes of the first and second BA should begin first by studying a series of isolated defects (cleft lip with or without CP, micrognathia, and EAC atresia) that compose the major features of these syndromes and allow a more specific diagnosis. After discussion of these defects and the associated embryology, we discuss PRS, HFM, ACS, TCS, Stickler syndrome, and VCFS.

Syndromes of the first and second branchial arches, part 1: embryology and characteristic defects.

A variety of congenital syndromes affecting the face occur due to defects involving the first and second BAs. Radiographic evaluation of craniofacial deformities is necessary to define aberrant anatomy, plan surgical procedures, and evaluate the effects of craniofacial growth and surgical reconstructions. High-resolution CT has proved vital in determining the nature and extent of these syndromes. The radiologic evaluation of syndromes of the first and second BAs should begin first by studying a series of isolated defects: CL with or without CP, micrognathia, and EAC atresia, which compose the major features of these syndromes and allow more specific diagnosis. After discussion of these defects and the associated embryology, we proceed to discuss the VCFS, PRS, ACS, TCS, Stickler syndrome, and HFM.

A venous cause for facial canal enlargement: multidetector row CT findings and histopathologic correlation.

An enlarged facial nerve canal can be a seen in both pathologic and nonpathologic processes. The purposes of this report are the following: 1) to present a rare cause of bony facial nerve canal enlargement, due to an enlarged vein, with high-resolution MDCT and histopathologic correlation; and 2) to discuss the vascular anatomy that gives rise to this variant.

Optic neuritis: imaging with magnetization transfer.

OBJECTIVE: Our objective was to prospectively examine the optic nerves in patients with clinically severe unilateral optic neuritis, using routine spin-echo and magnetization transfer MR imaging. SUBJECTS AND METHODS: For 39 patients with such lesions, we calculated the magnetization transfer ratio along the involved intraorbital optic nerve and along the asymptomatic contralateral optic nerve in a mirror-image location. Magnetization transfer ratios were correlated with the imaging findings on routine spin-echo MR imaging. RESULTS: Magnetization transfer ratios were decreased in 33 of the 39 clinically symptomatic optic nerves, including 12 of the 18 clinically symptomatic optic nerves in which no abnormality was seen on routine spin-echo MR images obtained before and after administration of gadopentetate dimeglumine. CONCLUSION: Magnetization transfer imaging reveals intraorbital optic nerve abnormalities in patients with optic neuritis even when such lesions are otherwise occult on routine magnetization transfer imaging.

Simultaneous bilateral internal carotid and vertebral artery dissection following chiropractic manipulation: case report and review of the literature.

Single-vessel cervical arterial dissections typically occur in young adults and are a common cause of cerebral ischemia and stroke. Although the pathogenesis of multivessel dissection is unclear, it is thought to be a consequence of underlying collagen vascular disease. We present a 34-year-old previously healthy man who developed bilateral internal carotid and vertebral artery dissection following chiropractic manipulation.