COVID-19 mortality prediction in the intensive care unit with deep learning based on longitudinal chest X-rays and clinical data.

OBJECTIVES: We aimed to develop deep learning models using longitudinal chest X-rays (CXRs) and clinical data to predict in-hospital mortality of COVID-19 patients in the intensive care unit (ICU). METHODS: Six hundred fifty-four patients (212 deceased, 442 alive, 5645 total CXRs) were identified across two institutions. Imaging and clinical data from one institution were used to train five longitudinal transformer-based networks applying five-fold cross-validation. The models were tested on data from the other institution, and pairwise comparisons were used to determine the best-performing models. RESULTS: A higher proportion of deceased patients had elevated white blood cell count, decreased absolute lymphocyte count, elevated creatine concentration, and incidence of cardiovascular and chronic kidney disease. A model based on pre-ICU CXRs achieved an AUC of 0.632 and an accuracy of 0.593, and a model based on ICU CXRs achieved an AUC of 0.697 and an accuracy of 0.657. A model based on all longitudinal CXRs (both pre-ICU and ICU) achieved an AUC of 0.702 and an accuracy of 0.694. A model based on clinical data alone achieved an AUC of 0.653 and an accuracy of 0.657. The addition of longitudinal imaging to clinical data in a combined model significantly improved performance, reaching an AUC of 0.727 (p = 0.039) and an accuracy of 0.732. CONCLUSIONS: The addition of longitudinal CXRs to clinical data significantly improves mortality prediction with deep learning for COVID-19 patients in the ICU. KEY POINTS: • Deep learning was used to predict mortality in COVID-19 ICU patients. • Serial radiographs and clinical data were used. • The models could inform clinical decision-making and resource allocation.

Interocular Difference in Retinal Nerve Fiber Layer Thickness Predicts Optic Neuritis in Pediatric-Onset Multiple Sclerosis.

BACKGROUND: Optical coherence tomography (OCT) is capable of quantifying retinal damage. Defining the extent of anterior visual pathway injury is important in multiple sclerosis (MS) as a way to document evidence of prior disease, including subclinical injury, and setting a baseline for patients early in the course of disease. Retinal nerve fiber layer (RNFL) thickness is typically classified as low if values fall outside of a predefined range for a healthy population. In adults, an interocular difference (IOD) in RNFL thickness greater than 5 µm identified a history of unilateral optic neuritis (ON). Through our PERCEPTION (PEdiatric Research Collaboration ExPloring Tests in Ocular Neuroimmunology) study, we explored whether RNFL IOD informs on remote ON in a multicenter pediatric-onset MS (POMS) cohort. METHODS: POMS (defined using consensus criteria and first attack <18 years) patients were recruited from 4 academic centers. A clinical history of ON (>6 months prior to an OCT scan) was confirmed by medical record review. RNFL thickness was measured on Spectralis machines (Heidelberg, Germany). Using a cohort of healthy controls from our centers tested on the same machines, RNFL thickness <86 µm (<2 SDs below the mean) was defined as abnormal. Based on previously published findings in adults, an RNFL IOD >5 µm was defined as abnormal. The proportions of POMS participants with RNFL thinning (<86 µm) and abnormal IOD (>5 µm) were calculated. Logistic regression was used to determine whether IOD was associated with remote ON. RESULTS: A total of 157 participants with POMS (mean age 15.2 years, SD 3.2; 67 [43%] with remote ON) were enrolled. RNFL thinning occurred in 45 of 90 (50%) ON eyes and 24 of 224 (11%) non-ON eyes. An IOD >5 µm was associated with a history of remote ON (P < 0.001). An IOD >5 µm occurred in 62 participants, 40 (65%) with remote ON. Among 33 participants with remote ON but normal RNFL values (≥86 µm in both eyes), 14 (42%) were confirmed to have ON by IOD criteria (>5 µm). CONCLUSIONS: In POMS, the diagnostic yield of OCT in confirming remote ON is enhanced by considering RNFL IOD, especially for those patients with RNFL thickness for each eye in the normal range. An IOD >5 µm in patients with previous visual symptoms suggests a history of remote ON.

Structural correlates of atypical visual and motor cortical oscillations in pediatric-onset multiple sclerosis.

We have previously demonstrated that pediatric-onset multiple sclerosis (POMS) negatively impacts the visual pathway as well as motor processing speed. Relationships between MS-related diffuse structural damage of gray and white matter (WM) tissue and cortical responses to visual and motor stimuli remain poorly understood. We used magnetoencephalography in 14 POMS patients and 15 age- and sex-matched healthy controls to assess visual gamma (30-80 Hz), motor gamma (60-90 Hz), and motor beta (15-30 Hz) cortical oscillatory responses to a visual-motor task. Then, 3T MRI was used to: (a) calculate fractional anisotropy (FA) of the posterior visual and corticospinal motor WM pathways and (b) quantify volume and thickness of the cuneus and primary motor cortex. Visual gamma band power was reduced in POMS and was associated with reduced FA of the optic radiations but not with loss of cuneus volume or thickness. Activity in the primary motor cortex, as measured by postmovement beta rebound amplitude associated with peak latency, was decreased in POMS, although this reduction was not predicted by structural metrics. Our findings implicate loss of WM integrity as a contributor to reduced electrical responses in the visual cortex in POMS. Future work in larger cohorts will inform on the cognitive implications of this finding in terms of visual processing function and will determine whether the progressive loss of brain volume known to occur in POMS ultimately contributes to both progressive dysfunction in such tasks as well as progressive reduction in cortical electrical responses in the visual cortex.

Multi-modality Regional Alignment Network for Covid X-Ray Survival Prediction and Report Generation.

In response to the worldwide COVID-19 pandemic, advanced automated technologies have emerged as valuable tools to aid healthcare professionals in managing an increased workload by improving radiology report generation and prognostic analysis. This study proposes a Multi-modality Regional Alignment Network (MRANet), an explainable model for radiology report generation and survival prediction that focuses on high-risk regions. By learning spatial correlation in the detector, MRANet visually grounds region-specific descriptions, providing robust anatomical regions with a completion strategy. The visual features of each region are embedded using a novel survival attention mechanism, offering spatially and risk-aware features for sentence encoding while maintaining global coherence across tasks. A cross-domain LLMs-Alignment is employed to enhance the image-to-text transfer process, resulting in sentences rich with clinical detail and improved explainability for radiologists. Multi-center experiments validate the overall performance and each module's composition within the model, encouraging further advancements in radiology report generation research emphasizing clinical interpretation and trustworthiness in AI models applied to medical studies.

Acquisition and Loss of Developmental Milestones and Time to Disease-Related Outcomes in Cerebral Alexander Disease.

Objective: To determine the ages at acquisition of developmental milestones, loss of motor function, and clinical symptoms in Alexander disease. Methods: Patients with confirmed cerebral Alexander disease were included. Data abstraction of developmental and disease-specific milestones was performed from medical records, physical exams, and questionnaires. Mixed effects logistic regression was used to determine if key clinical features were associated with milestone achievement, controlling for patient age. Results: 51 patients with cerebral/infantile Alexander disease were evaluated at a mean age of 10.96 years (range 2.29-31.08 years). Developmental milestones in Alexander disease were often achieved but delayed. Ambulation was achieved in 44 subjects (86%); 34 (67%) subjects walked independently (mean age 1.9 years, range 0.91-3.25 years) and an additional 10 (20%) subjects walked with assistance (mean age 3.9 years, range 1.8-8 years) but did not progress to independent ambulation. Developmental delay was the earliest and most prevalent symptom (N = 48 [94%], mean age 0.58 years), compared to an initial seizure (N = 41 [80%], mean age 2.80 years), and macrocephaly (N = 28 [55%], mean age 4.04 years), P < .0001 between these ages of onset. Loss of independent ambulation occurred in 11 of the 34 (32%) children who had acquired ambulation (range 3.41-15.10 years). Presence of seizures or macrocephaly did not predict the achievement or loss of ambulation. Conclusions: The clinical triad of developmental delay, seizures, and macrocephaly are not universally present in cerebral Alexander disease. Clinicians should have a high index of suspicion for Alexander disease in patients with mild delays and a first seizure.

Artificial intelligence for medical image analysis in epilepsy.

Given improvements in computing power, artificial intelligence (AI) with deep learning has emerged as the state-of-the art method for the analysis of medical imaging data and will increasingly be used in the clinical setting. Recent work in epilepsy research has aimed to use AI methods to improve diagnosis, prognosis, and treatment, with the ultimate goal of developing highly accurate and reliable tools to aid clinical decision making. Here, we review how researchers are currently using AI methods in the analysis of neuroimaging data in epilepsy, focusing on challenges unique to each imaging modality with an emphasis on clinical significance. We further provide critical analyses of existing techniques and recommend areas for future work. We call for: (1) a multimodal approach that leverages the strengths of different modalities while compensating for their individual weaknesses, and (2) widespread implementation of generalizability testing of proposed models, a needed step before their introduction into clinical workflows. To achieve both goals, more collaborations among research groups and institutions in this field will be required.

Prediction of lung malignancy progression and survival with machine learning based on pre-treatment FDG-PET/CT.

BACKGROUND: Pre-treatment FDG-PET/CT scans were analyzed with machine learning to predict progression of lung malignancies and overall survival (OS). METHODS: A retrospective review across three institutions identified patients with a pre-procedure FDG-PET/CT and an associated malignancy diagnosis. Lesions were manually and automatically segmented, and convolutional neural networks (CNNs) were trained using FDG-PET/CT inputs to predict malignancy progression. Performance was evaluated using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Image features were extracted from CNNs and by radiomics feature extraction, and random survival forests (RSF) were constructed to predict OS. Concordance index (C-index) and integrated brier score (IBS) were used to evaluate OS prediction. FINDINGS: 1168 nodules (n=965 patients) were identified. 792 nodules had progression and 376 were progression-free. The most common malignancies were adenocarcinoma (n=740) and squamous cell carcinoma (n=179). For progression risk, the PET+CT ensemble model with manual segmentation (accuracy=0.790, AUC=0.876) performed similarly to the CT only (accuracy=0.723, AUC=0.888) and better compared to the PET only (accuracy=0.664, AUC=0.669) models. For OS prediction with deep learning features, the PET+CT+clinical RSF ensemble model (C-index=0.737) performed similarly to the CT only (C-index=0.730) and better than the PET only (C-index=0.595), and clinical only (C-index=0.595) models. RSF models constructed with radiomics features had comparable performance to those with CNN features. INTERPRETATION: CNNs trained using pre-treatment FDG-PET/CT and extracted performed well in predicting lung malignancy progression and OS. OS prediction performance with CNN features was comparable to a radiomics approach. The prognostic models could inform treatment options and improve patient care. FUNDING: NIH NHLBI training grant (5T35HL094308-12, John Sollee).

Deep learning-based automatic tumor burden assessment of pediatric high-grade gliomas, medulloblastomas, and other leptomeningeal seeding tumors.

BACKGROUND: Longitudinal measurement of tumor burden with magnetic resonance imaging (MRI) is an essential component of response assessment in pediatric brain tumors. We developed a fully automated pipeline for the segmentation of tumors in pediatric high-grade gliomas, medulloblastomas, and leptomeningeal seeding tumors. We further developed an algorithm for automatic 2D and volumetric size measurement of tumors. METHODS: The preoperative and postoperative cohorts were randomly split into training and testing sets in a 4:1 ratio. A 3D U-Net neural network was trained to automatically segment the tumor on T1 contrast-enhanced and T2/FLAIR images. The product of the maximum bidimensional diameters according to the RAPNO (Response Assessment in Pediatric Neuro-Oncology) criteria (AutoRAPNO) was determined. Performance was compared to that of 2 expert human raters who performed assessments independently. Volumetric measurements of predicted and expert segmentations were computationally derived and compared. RESULTS: A total of 794 preoperative MRIs from 794 patients and 1003 postoperative MRIs from 122 patients were included. There was excellent agreement of volumes between preoperative and postoperative predicted and manual segmentations, with intraclass correlation coefficients (ICCs) of 0.912 and 0.960 for the 2 preoperative and 0.947 and 0.896 for the 2 postoperative models. There was high agreement between AutoRAPNO scores on predicted segmentations and manually calculated scores based on manual segmentations (Rater 2 ICC = 0.909; Rater 3 ICC = 0.851). Lastly, the performance of AutoRAPNO was superior in repeatability to that of human raters for MRIs with multiple lesions. CONCLUSIONS: Our automated deep learning pipeline demonstrates potential utility for response assessment in pediatric brain tumors. The tool should be further validated in prospective studies.

An automated COVID-19 triage pipeline using artificial intelligence based on chest radiographs and clinical data.

While COVID-19 diagnosis and prognosis artificial intelligence models exist, very few can be implemented for practical use given their high risk of bias. We aimed to develop a diagnosis model that addresses notable shortcomings of prior studies, integrating it into a fully automated triage pipeline that examines chest radiographs for the presence, severity, and progression of COVID-19 pneumonia. Scans were collected using the DICOM Image Analysis and Archive, a system that communicates with a hospital's image repository. The authors collected over 6,500 non-public chest X-rays comprising diverse COVID-19 severities, along with radiology reports and RT-PCR data. The authors provisioned one internally held-out and two external test sets to assess model generalizability and compare performance to traditional radiologist interpretation. The pipeline was evaluated on a prospective cohort of 80 radiographs, reporting a 95% diagnostic accuracy. The study mitigates bias in AI model development and demonstrates the value of an end-to-end COVID-19 triage platform.

Effect of age and neurofibromatosis type 1 status on white matter integrity in the optic radiations.

BACKGROUND: Adults with neurofibromatosis type 1 (NF1) have decreased white matter integrity, but differences in children with NF1 have not been described. Defining normal values for diffusion tensor imaging (DTI) measures, especially in the optic radiations, is important to the development of DTI as a potential biomarker of visual acuity in children with optic pathway glioma. This study examines the effect of age and NF1 status on DTI measures in children. METHODS: In this retrospective study, MR imaging including DTI was conducted in 93 children (40 children with NF1 and 53 healthy controls) between 0 and 14 years of age. Regression models of age, sex, and NF1 status on DTI measures were evaluated, and tract-based spatial statistics (TBSS) compared DTI measures in age-matched NF1 to non-NF1 cohorts. RESULTS: Fractional anisotropy, radial diffusivity, and mean diffusivity in white matter tracts of the optic radiations varied with age and were best modeled by a logarithmic function. Age-related DTI measure change was different in NF1 versus non-NF1 subjects. Normal values and 95% confidence intervals for age 0.5-12 years were derived for both groups. Differences in DTI measures between NF1 and non-NF1 groups at a range of ages were shown diffusely throughout the cerebral white matter using TBSS. CONCLUSIONS: Children with NF1 demonstrate increased diffusion throughout the brain compared to children without NF1 suggesting a potentially altered developmental trajectory of optic radiation microstructure. Defining normal values for white matter integrity in children with NF1 may help target early intervention efforts in this vulnerable group.

Effects of Optic Neuritis, T2 Lesions, and Microstructural Diffusion Integrity in the Visual Pathway on Cortical Thickness in Pediatric-Onset Multiple Sclerosis.

BACKGROUND AND PURPOSE: Pediatric-onset multiple sclerosis (POMS) is associated with focal inflammatory lesions and the loss of cortical and deep gray matter. Optic neuritis (ON) and white matter (WM) lesions in the visual pathway can directly contribute to visual cortical mantle thinning. We determine the relative contributions of MS insult on anterior and posterior visual pathway integrity. METHODS: High- and low-contrast visual acuity, optical coherence tomography (OCT), and 3T MRI scans were obtained from 20 POMS patients (10 with remote ON) and 22 age- and sex-matched healthy controls. Cortical mantle thickness was measured using FreeSurfer. Fractional anisotropy (FA) and mean diffusivity were calculated for postchiasmal optic radiations (with and without WM lesions). Groups were compared using Student's t-test (adjusted for multiple comparisons), and simple linear regression was used to investigate interrelationships between measures. RESULTS: Mean cortical thickness of the whole brain was reduced in patients (2.49 mm) versus controls (2.58 mm, P = .0432) and in the visual cortex (2.07 mm vs. 2.17 mm, P = .0059), although the foveal confluence was spared. Mean FA of the optic radiations was reduced in POMS (.40) versus controls (.43, P = .0042) and correlated with visual cortical mantle thickness in POMS (P = .017). Visual acuity, OCT measures, and lesion volumes in the optic radiations were not associated with cortical mantle thickness. CONCLUSIONS: POMS negatively impacts the integrity of the anterior visual pathway, but it is the loss of WM integrity that drives anterograde loss of the cortical mantle. Preserved visual acuity and foveal sparing imply some degree of functional and structural resilience.