Multi-modal trained artificial intelligence solution to triage chest X-ray for COVID-19 using pristine ground-truth, versus radiologists.

The front-line imaging modalities computed tomography (CT) and X-ray play important roles for triaging COVID patients. Thoracic CT has been accepted to have higher sensitivity than a chest X-ray for COVID diagnosis. Considering the limited access to resources (both hardware and trained personnel) and issues related to decontamination, CT may not be ideal for triaging suspected subjects. Artificial intelligence (AI) assisted X-ray based application for triaging and monitoring require experienced radiologists to identify COVID patients in a timely manner with the additional ability to delineate and quantify the disease region is seen as a promising solution for widespread clinical use. Our proposed solution differs from existing solutions presented by industry and academic communities. We demonstrate a functional AI model to triage by classifying and segmenting a single chest X-ray image, while the AI model is trained using both X-ray and CT data. We report on how such a multi-modal training process improves the solution compared to single modality (X-ray only) training. The multi-modal solution increases the AUC (area under the receiver operating characteristic curve) from 0.89 to 0.93 for a binary classification between COVID-19 and non-COVID-19 cases. It also positively impacts the Dice coefficient (0.59 to 0.62) for localizing the COVID-19 pathology. To compare the performance of experienced readers to the AI model, a reader study is also conducted. The AI model showed good consistency with respect to radiologists. The DICE score between two radiologists on the COVID group was 0.53 while the AI had a DICE value of 0.52 and 0.55 when compared to the segmentation done by the two radiologists separately. From a classification perspective, the AUCs of two readers was 0.87 and 0.81 while the AUC of the AI is 0.93 based on the reader study dataset. We also conducted a generalization study by comparing our method to the-state-art methods on independent datasets. The results show better performance from the proposed method. Leveraging multi-modal information for the development benefits the single-modal inferencing.

Longitudinal Resting State Functional Connectivity Predicts Clinical Outcome in Mild Traumatic Brain Injury.

Mild traumatic brain injury (mTBI) affects about 42 million people worldwide. It is often associated with headache, cognitive deficits, and balance difficulties but rarely shows any abnormalities on conventional computed tomography (CT) or magnetic resonance imaging (MRI). Although in most mTBI patients the symptoms resolve within 3 months, 10-15% of patients continue to exhibit symptoms beyond a year. Also, it is known that there exists a vulnerable period post-injury, when a second injury may exacerbate clinical prognosis. Identifying this vulnerable period may be critical for patient outcome, but very little is known about the neural underpinnings of mTBI and its recovery. In this work, we used advanced functional neuroimaging to study longitudinal changes in functional organization of the brain during the 3-month recovery period post-mTBI. Fractional amplitude of low frequency fluctuations (fALFF) measured from resting state functional MRI (rs-fMRI) was found to be associated with symptom severity score (SSS, r = -0.28, p = 0.002). Decreased fALFF was observed in specific functional networks for patients with higher SSS, and fALFF returned to higher values when the patient recovered (lower SSS). In addition, functional connectivity of the same networks was found to be associated with concurrent SSS, and connectivity immediately after injury (<10 days) was capable of predicting SSS at a later time-point (3 weeks to 3 months, p < 0.05). Specific networks including motor, default-mode, and visual networks were found to be associated with SSS (p < 0.001), and connectivity between these networks predicted 3-month clinical outcome (motor and visual: p < 0.001, default-mode: p < 0.006). Our results suggest that functional connectivity in these networks comprise potential biomarkers for predicting mTBI recovery profiles and clinical outcome.

Recursive feature elimination for biomarker discovery in resting-state functional connectivity.

Biomarker discovery involves finding correlations between features and clinical symptoms to aid clinical decision. This task is especially difficult in resting state functional magnetic resonance imaging (rs-fMRI) data due to low SNR, high-dimensionality of images, inter-subject and intra-subject variability and small numbers of subjects compared to the number of derived features. Traditional univariate analysis suffers from the problem of multiple comparisons. Here, we adopt an alternative data-driven method for identifying population differences in functional connectivity. We propose a machine-learning approach to down-select functional connectivity features associated with symptom severity in mild traumatic brain injury (mTBI). Using this approach, we identified functional regions with altered connectivity in mTBI. including the executive control, visual and precuneus networks. We compared functional connections at multiple resolutions to determine which scale would be more sensitive to changes related to patient recovery. These modular network-level features can be used as diagnostic tools for predicting disease severity and recovery profiles.

Simulations of virtual PET/CT 3-D bronchoscopy imaging using a physical porcine lung-heart phantom.

PURPOSE: We present a systematic approach for studying positron emission tomography-computed tomography (PET/CT) 3-D virtual fly-through endoscopy and for assessing the accuracy of this technology for visualizing and detecting endobronchial lesions as a function of focal lesion morphology and activity. PROCEDURES: Capsules designed to simulate endobronchial lesions were filled with activity and introduced into a porcine lung-heart phantom. PET/CT images were acquired, reconstructed, and volume rendered as 3-D fly-through and fly-around visualizations. Anatomical positioning of lesions seen on the 3-D-volume-rendered PET/CT images was compared to the actual position of the capsules. RESULTS: Lesion size was observed to be highly sensitive to PET threshold parameter settings and careful opacity and color transfer function parameter assignment. CONCLUSION: We have demonstrated a phantom model for studies of PET/CT 3-D virtual fly-through bronchoscopy and have applied this model for understanding the effect of PET thresholding on the visualization and detection of lesions.

A new method for the study of velopharyngeal function using gated magnetic resonance imaging.

The purpose of this project was to assess the feasibility of imaging the velopharynx of adult volunteers during repetitive speech, using gated magnetic resonance imaging (MRI). Although a number of investigators have used conventional MRI in the study of the human vocal tract, the mismatch between the lengthy time necessary to acquire sufficiently detailed images and the rapidity of movement of the vocal tract during speech has forced investigators to acquire images either while the subject is at rest or during sustained utterances. The technique used here acquired a portion of each image during repetitive utterances, building the full image over multiple utterance cycles. The velopharyngeal portal was imaged on a 1.5-Tesla GE Signa LX 8.2 platform with gated fast spoiled gradient echo protocol. An external 1-Hertz trigger was fed to the cardiac gate. Subjects synchronized utterance of consonant-vowel syllables to a flashing light synchronized with the external trigger. Each acquisition of 30 phases per second at a single-slice location took 22 to 29 seconds. Four consonant-vowel syllables (/pa/, /ma/, /sa/, and /ka/) were evaluated. Subjects vocalized throughout the acquisition, beginning 5 to 6 seconds beforehand to establish a regular rhythm. Imaging of the velopharyngeal portal was performed for sagittal, velopharyngeal axial (aligned perpendicular to the "knee" of the velum), axial, and coronal planes. Volumes were obtained by sequential acquisition of six to 10 slices (each with 30 phases) in the axial or sagittal planes during repetition of the /pa/ syllable. Spatiotemporal volumes of the single-slice data were sectioned to provide time-motion images (analogous to M-mode echocardiograms). Three-dimensional dynamic volume renderings of palate motion were displayed interactively (Vortex; CieMed, Singapore). A method suitable for the collection and visualization of four-dimensional information regarding monosyllabic speech using gated MRI was developed. These techniques were applied to a population of adult volunteer subjects with no history of speech problems and two patients with a history of cleft lip and palate. The techniques allowed good real-time visualization of velopharyngeal anatomy during its entire range of motion and was also able to image pathology-specific anatomic differences in the subjects with cleft lip and cleft palate. These methods may be applicable to a wide spectrum of problems in speech physiology research and for clinical decision-making regarding surgery for speech and outcomes analysis.