Deep Learning Detection of Active Pulmonary Tuberculosis at Chest Radiography Matched the Clinical Performance of Radiologists.

Background The World Health Organization (WHO) recommends chest radiography to facilitate tuberculosis (TB) screening. However, chest radiograph interpretation expertise remains limited in many regions. Purpose To develop a deep learning system (DLS) to detect active pulmonary TB on chest radiographs and compare its performance to that of radiologists. Materials and Methods A DLS was trained and tested using retrospective chest radiographs (acquired between 1996 and 2020) from 10 countries. To improve generalization, large-scale chest radiograph pretraining, attention pooling, and semisupervised learning ("noisy-student") were incorporated. The DLS was evaluated in a four-country test set (China, India, the United States, and Zambia) and in a mining population in South Africa, with positive TB confirmed with microbiological tests or nucleic acid amplification testing (NAAT). The performance of the DLS was compared with that of 14 radiologists. The authors studied the efficacy of the DLS compared with that of nine radiologists using the Obuchowski-Rockette-Hillis procedure. Given WHO targets of 90% sensitivity and 70% specificity, the operating point of the DLS (0.45) was prespecified to favor sensitivity. Results A total of 165 754 images in 22 284 subjects (mean age, 45 years; 21% female) were used for model development and testing. In the four-country test set (1236 subjects, 17% with active TB), the receiver operating characteristic (ROC) curve of the DLS was higher than those for all nine India-based radiologists, with an area under the ROC curve of 0.89 (95% CI: 0.87, 0.91). Compared with these radiologists, at the prespecified operating point, the DLS sensitivity was higher (88% vs 75%, P < .001) and specificity was noninferior (79% vs 84%, P = .004). Trends were similar within other patient subgroups, in the South Africa data set, and across various TB-specific chest radiograph findings. In simulations, the use of the DLS to identify likely TB-positive chest radiographs for NAAT confirmation reduced the cost by 40%-80% per TB-positive patient detected. Conclusion A deep learning method was found to be noninferior to radiologists for the determination of active tuberculosis on digital chest radiographs. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by van Ginneken in this issue.

Neural dynamics necessary and sufficient for transition into pre-sleep induced by EEG neurofeedback.

The transition from being fully awake to pre-sleep occurs daily just before falling asleep; thus its disturbance might be detrimental. Yet, the neuronal correlates of the transition remain unclear, mainly due to the difficulty in capturing its inherent dynamics. We used an EEG theta/alpha neurofeedback to rapidly induce the transition into pre-sleep and simultaneous fMRI to reveal state-dependent neural activity. The relaxed mental state was verified by the corresponding enhancement in the parasympathetic response. Neurofeedback sessions were categorized as successful or unsuccessful, based on the known EEG signature of theta power increases over alpha, temporally marked as a distinct "crossover" point. The fMRI activation was considered before and after this point. During successful transition into pre-sleep the period before the crossover was signified by alpha modulation that corresponded to decreased fMRI activity mainly in sensory gating related regions (e.g. medial thalamus). In parallel, although not sufficient for the transition, theta modulation corresponded with increased activity in limbic and autonomic control regions (e.g. hippocampus, cerebellum vermis, respectively). The post-crossover period was designated by alpha modulation further corresponding to reduced fMRI activity within the anterior salience network (e.g. anterior cingulate cortex, anterior insula), and in contrast theta modulation corresponded to the increased variance in the posterior salience network (e.g. posterior insula, posterior cingulate cortex). Our findings portray multi-level neural dynamics underlying the mental transition from awake to pre-sleep. To initiate the transition, decreased activity was required in external monitoring regions, and to sustain the transition, opposition between the anterior and posterior parts of the salience network was needed, reflecting shifting from extra- to intrapersonal based processing, respectively.

Deep learning for distinguishing normal versus abnormal chest radiographs and generalization to two unseen diseases tuberculosis and COVID-19.

Chest radiography (CXR) is the most widely-used thoracic clinical imaging modality and is crucial for guiding the management of cardiothoracic conditions. The detection of specific CXR findings has been the main focus of several artificial intelligence (AI) systems. However, the wide range of possible CXR abnormalities makes it impractical to detect every possible condition by building multiple separate systems, each of which detects one or more pre-specified conditions. In this work, we developed and evaluated an AI system to classify CXRs as normal or abnormal. For training and tuning the system, we used a de-identified dataset of 248,445 patients from a multi-city hospital network in India. To assess generalizability, we evaluated our system using 6 international datasets from India, China, and the United States. Of these datasets, 4 focused on diseases that the AI was not trained to detect: 2 datasets with tuberculosis and 2 datasets with coronavirus disease 2019. Our results suggest that the AI system trained using a large dataset containing a diverse array of CXR abnormalities generalizes to new patient populations and unseen diseases. In a simulated workflow where the AI system prioritized abnormal cases, the turnaround time for abnormal cases reduced by 7-28%. These results represent an important step towards evaluating whether AI can be safely used to flag cases in a general setting where previously unseen abnormalities exist. Lastly, to facilitate the continued development of AI models for CXR, we release our collected labels for the publicly available dataset.

Machine learning fMRI classifier delineates subgroups of schizophrenia patients.

BACKGROUND: The search for a validated neuroimaging-based brain marker in psychiatry has thus far been fraught with both clinical and methodological difficulties. The present study aimed to apply a novel data-driven machine-learning approach to functional Magnetic Resonance Imaging (fMRI) data obtained during a cognitive task in order to delineate the neural mechanisms involved in two schizophrenia subgroups: schizophrenia patients with and without Obsessive-Compulsive Disorder (OCD). METHODS: 16 schizophrenia patients with OCD ("schizo-obsessive"), 17 pure schizophrenia patients, and 20 healthy controls underwent fMRI while performing a working memory task. A whole brain search for activation clusters of cognitive load was performed using a recently developed data-driven multi-voxel pattern analysis (MVPA) approach, termed Searchlight Based Feature Extraction (SBFE), and which yields a robust fMRI-based classifier. RESULTS: The SBFE successfully classified the two schizophrenia groups with 91% accuracy based on activations in the right intraparietal sulcus (r-IPS), which further correlated with reduced symptom severity among schizo-obsessive patients. CONCLUSIONS: The results indicate that this novel SBFE approach can successfully delineate between symptom dimensions in the context of complex psychiatric morbidity.

Enhanced functional synchronization of medial and lateral PFC underlies internally-guided action planning.

Actions are often internally guided, reflecting our covert will and intentions. The dorsomedial prefrontal cortex, including the pre-Supplementary Motor Area (pre-SMA), has been implicated in the internally generated aspects of action planning, such as choice and intention. Yet, the mechanism by which this area interacts with other cognitive brain regions such as the dorsolateral prefrontal cortex, a central node in decision-making, is still unclear. To shed light on this mechanism, brain activity was measured via fMRI and intracranial EEG in two studies during the performance of visually cued repeated finger tapping in which the choice of finger was guided by either a presented number (external) or self-choice (internal). A functional-MRI (fMRI) study in 15 healthy participants demonstrated that the pre-SMA, compared to the SMA proper, was more active and also more functionally correlated with the dorsolateral prefrontal cortex during internally compared to externally guided action planning (p < 0.05, random effect). In a similar manner, an intracranial-EEG study in five epilepsy patients showed greater inter-regional gamma-related connectivity between electrodes situated in medial and lateral aspects of the prefrontal cortex for internally compared to externally guided actions. Although this finding was observed for groups of electrodes situated both in the pre-SMA and SMA-proper, increased intra-cluster gamma-related connectivity was only observed for the pre-SMA (sign-test, p < 0.0001). Overall our findings provide multi-scale indications for the involvement of the dorsomedial prefrontal cortex, and especially the pre-SMA, in generating internally guided motor planning. Our intracranial-EEG results further point to enhanced functional connectivity between decision-making- and motor planning aspects of the PFC, as a possible neural mechanism for internally generated action planning.