AIforCOVID: Predicting the clinical outcomes in patients with COVID-19 applying AI to chest-X-rays. An Italian multicentre study.

Recent epidemiological data report that worldwide more than 53 million people have been infected by SARS-CoV-2, resulting in 1.3 million deaths. The disease has been spreading very rapidly and few months after the identification of the first infected, shortage of hospital resources quickly became a problem. In this work we investigate whether artificial intelligence working with chest X-ray (CXR) scans and clinical data can be used as a possible tool for the early identification of patients at risk of severe outcome, like intensive care or death. Indeed, further to induce lower radiation dose than computed tomography (CT), CXR is a simpler and faster radiological technique, being also more widespread. In this respect, we present three approaches that use features extracted from CXR images, either handcrafted or automatically learnt by convolutional neuronal networks, which are then integrated with the clinical data. As a further contribution, this work introduces a repository that collects data from 820 patients enrolled in six Italian hospitals in spring 2020 during the first COVID-19 emergency. The dataset includes CXR images, several clinical attributes and clinical outcomes. Exhaustive evaluation shows promising performance both in 10-fold and leave-one-centre-out cross-validation, suggesting that clinical data and images have the potential to provide useful information for the management of patients and hospital resources.

Lung Ultrasound in the Diagnosis of COVID-19 Pneumonia: Not Always and Not Only What Is COVID-19 "Glitters".

Background: In the current coronavirus disease-2019 (COVID-19) pandemic, lung ultrasound (LUS) has been extensively employed to evaluate lung involvement and proposed as a useful screening tool for early diagnosis in the emergency department (ED), prehospitalization triage, and treatment monitoring of COVID-19 pneumonia. However, the actual effectiveness of LUS in characterizing lung involvement in COVID-19 is still unclear. Our aim was to evaluate LUS diagnostic performance in assessing or ruling out COVID-19 pneumonia when compared with chest CT (gold standard) in a population of SARS-CoV-2-infected patients. Methods: A total of 260 consecutive RT-PCR confirmed SARS-CoV-2-infected patients were included in the study. All the patients underwent both chest CT scan and concurrent LUS at admission, within the first 6-12 h of hospital stay. Results: Chest CT scan was considered positive when showing a "typical" or "indeterminate" pattern for COVID-19, according to the RSNA classification system. Disease prevalence for COVID-19 pneumonia was 90.77%. LUS demonstrated a sensitivity of 56.78% in detecting lung alteration. The concordance rate for the assessment of abnormalities by both methods increased in the case of peripheral distribution and middle-lower lung location of lesions and in cases of more severe lung involvement. A total of nine patients had a "false-positive" LUS examination. Alternative diagnosis included chronic heart disease (six cases), bronchiectasis (two cases), and subpleural emphysema (one case). LUS specificity was 62.50%. Collateral findings indicative of overlapping conditions at chest CT were recorded also in patients with COVID-19 pneumonia and appeared distributed with increasing frequency passing from the group with mild disease (17 cases) to that with severe disease (40 cases). Conclusions: LUS does not seem to be an adequate tool for screening purposes in the ED, due to the risk of missing some lesions and/or to underestimate the actual extent of the disease. Furthermore, the not specificity of LUS implies the possibility to erroneously classify pre-existing or overlapping conditions as COVID-19 pneumonia. It seems more safe to integrate a positive LUS examination with clinical, epidemiological, laboratory, and radiologic findings to suggest a "virosis." Viral testing confirmation is always required.

Low Sensitivity of Admission Lung US Compared to Chest CT for Diagnosis of Lung Involvement in a Cohort of 82 Patients with COVID-19 Pneumonia.

Background and Objectives: The potential role of lung ultrasound (LUS) in characterizing lung involvement in Coronavirus disease 2019 (COVID-19) is still debated. The aim of the study was to estimate sensitivity of admission LUS for the detection of SARS-CoV-2 lung involvement using Chest-CT (Computed Tomography) as reference standard in order to assess LUS usefulness in ruling out COVID-19 pneumonia in the Emergency Department (ED). Methods: Eighty-two patients with confirmed COVID-19 and signs of lung involvement on Chest-CT were consecutively admitted to our hospital and recruited in the study. Chest-CT and LUS examination were concurrently performed within the first 6-12h from admission. Sensitivity of LUS was calculated using CT findings as a reference standard. Results: Global LUS sensitivity in detecting COVID-19 pulmonary lesions was 52%. LUS sensitivity ranged from 8% in case of focal and sporadic ground-glass opacities (mild disease), to 52% for a crazy-paving pattern (moderate disease) and up to 100% in case of extensive subpleural consolidations (severe disease), although LUS was not always able to detect all the consolidations assessed at Chest-CT. LUS sensitivity was higher in detecting a typical Chest-CT pattern (60%) and abnormalities showing a middle-lower zone predominance (79%). Conclusions: As admission LUS may result falsely negative in most cases, it should not be considered as a reliable imaging tool in ruling out COVID-19 pneumonia in patients presenting in ED. It may at least represent an expanded clinical evaluation that needs integration with other diagnostic tests (e.g., nasopharyngeal swab, Chest-CT).

Diagnosis of COVID-19 in Patients with Negative Nasopharyngeal Swabs: Reliability of Radiological and Clinical Diagnosis and Accuracy Versus Serology.

BACKGROUND: The diagnosis of Coronavirus disease 2019 (COVID-19) relies on the positivity of nasopharyngeal swab. However, a significant percentage of symptomatic patients may test negative. We evaluated the reliability of COVID-19 diagnosis made by radiologists and clinicians and its accuracy versus serology in a sample of patients hospitalized for suspected COVID-19 with multiple negative swabs. METHODS: Admission chest CT-scans and clinical records of swab-negative patients, treated according to the COVID-19 protocol or deceased during hospitalization, were retrospectively evaluated by two radiologists and two clinicians, respectively. RESULTS: Of 254 patients, 169 swab-confirmed cases and one patient without chest CT-scan were excluded. A total of 84 patients were eligible for the reliability study. Of these, 21 patients died during hospitalization; the remaining 63 underwent serological testing and were eligible for the accuracy evaluation. Of the 63, 26 patients showed anti-Sars-Cov-2 antibodies, while 37 did not. The inter-rater agreement was "substantial" (kappa 0.683) between radiologists, "moderate" (kappa 0.454) between clinicians, and only "fair" (kappa 0.341) between radiologists and clinicians. Both radiologic and clinical evaluations showed good accuracy compared to serology. CONCLUSIONS: The radiologic and clinical diagnosis of COVID-19 for swab-negative patients proved to be sufficiently reliable and accurate to allow a diagnosis of COVID-19, which needs to be confirmed by serology and follow-up.