key: cord-0718386-vvk8gxe7
authors: Cau, Riccardo; Faa, Gavino; Nardi, Valentina; Balestrieri, Antonella; Puig, Josep; Suri, Jasjit S; SanFilippo, Roberto; Saba, Luca
title: Long-COVID diagnosis: from diagnostic to advanced AI-driven models
date: 2022-01-19
journal: Eur J Radiol
DOI: 10.1016/j.ejrad.2022.110164
sha: b23a01f9037fdde32d8c67cdaedd6759cddc2299
doc_id: 718386
cord_uid: vvk8gxe7

SARS-COV 2 is recognized to be responsible for a multi-organ syndrome. In most patients, symptoms are mild. However, in certain subjects, COVID-19 tends to progress more severely. Most of the patients infected with SARS-COV2 fully recovered within some weeks. In a considerable number of patients, like many other viral infections, various long-lasting symptoms have been described, now defined as “long COVID-19 syndrome”. Given the high number of contagious over the world, it is necessary to understand and comprehend this emerging pathology to enable early diagnosis and improve patents outcomes. In this scenario, AI-based models can be applied in long-COVID-19 patients to assist clinicians and at the same time, to reduce the considerable impact on the care and rehabilitation unit. The purpose of this manuscript is to review different aspects of long-COVID-19 syndrome from clinical presentation to diagnosis, highlighting the considerable impact that AI can have.

As of October 12 th , 2021, global casualties due to coronavirus disease 2019 (COVID-19) infections approach 200 million, several patients are asymptomatic or have mild symptoms, while few others show an aggressive and life-threatening disease [1] [2] [3] .

Through a large proportion of subjects infected with SARS-COV-2 present a restitutio ad integrum within a few weeks, an emerging aspect of COVID-19 infection is the long-term effect 4, 5 . Among those surviving the acute infection, a fraction of the COVID-19 patients reports persistent symptoms for 12 weeks or longer beyond the initial period of acute infection and illness 5 . The symptoms are diverse and related to multiorgan involvement, the most described symptoms are fatigue, headache, attention disorder, hair loss, and dyspnea (Figure 1) 6 In view of several reports, the definition of a new syndrome has been proposed, namely long COVID-19 syndrome. In the UK NICE guidelines, a distinction of long-COVID has been made in relation to the duration after the acute onset of the syndrome. Especially, post-acute COVID-19 is defined as ongoing symptomatic COVID-19 for patients who still have symptoms between 4 and 12 weeks after the start of acute symptoms, on the other hand, post-COVID-19 syndrome for people who still have symptoms for more than 12 weeks after the start of acute symptoms 8 .

Care and management for COVID-19 patients do not conclude after acute infection but continue in the outpatient setting. Consequently, a large amount of time and resources will be needed to help clinicians understand and manage long-term COVID-19 sequelae. In view of the multi-organ involvement during post-COVID-19 syndrome, an inter-disciplinary approach is mandatory, where clinical and laboratory data must be embedded with the radiological data ( Figure 2) . This scenario is a new future frontier for artificial intelligence (AI) such as its ability to join officebased, laboratory-based, and image-based data. We provide a review of the current literature on post-COVID-19 syndrome, its clinical presentation, and imaging-based diagnosis. Finally, we discuss the potential role of the multi-disciplinary approach with the support of AI.

The lung is a predominant organ involvement during COVID-19 infection 1 , but data on residual lung damage are still scarce. Among COVID-19 survivors several pulmonary manifestations, ranging from cough to dyspnea, have been described 9 13 . Further analysis showed the characteristic of patients with persistent lung injuries, including male, overweight, comorbidity, oxygen therapy, intensive care unit (ICU) admission, and invasive mechanical ventilation 14 .

Computed tomography (CT) imaging revealed pulmonary alteration in 63 % of survivors, including ground-glass opacities, consolidation, and reticulation with bilateral involvement in 75 % of survivors under analysis 12 .

The document from the European Society of Thoracic Imaging and the European Society of Radiology discussed the CT features in COVID-19 survivors at discharge and during the follow-up 15 .

The authors proposed a glossary of appropriate definitions to describe the lung abnormalities post-COVID-19 pneumonia, which include, for example, the term "fibrotic-like changes"

representing potential precursors of fibrosis, but with a high probability of resolving over time 15 . Given the high number of infected patients with SARS-COV2, persistent functional deficits described are likely to represent a significant disease burden. Therefore, prompt therapy may prevent potentially permanent fibrosis and functional impairment.

The British Thoracic Society provided guidance for the respiratory follow-up for patients with COVID-19 pneumonia in two separate algorithms based on the severity of acute COVID-19 infection and ICU admission required 16 . Both algorithms recommended clinical assessment, pulmonary function tests, and chest X-rays in all patients at 12 weeks. Subsequently, whether an abnormal chest X-ray and/or clinical impairment has been observed, a further high-resolution CT is recommended 16 . Also, NICE guidelines make recommendations for clinical investigation of patients with long-COVID, including chest X-rays in patients with persistent pulmonary symptoms 17 .

A review of evidence performed by the NICE described the presence of chest pain and palpitation in 10-44% survivors 18 . In an Italian case series, Carfi et al. reported the presence of chest pain in up to 20% of patients 60 days after acute infection 7 . Myocardial injuries have been described after SARS-COV 2 infection 19 . The psychological, social, and economic stress related to the COVID-19 was associated with an increased incidence of stress cardiomyopathy in comparison with prepandemic periods ( 7,8% vs 1,5-1,8%, respectively) 20 . Also, cardiac arrhythmias in Long-COVID patients were reported, including atrial fibrillation, supraventricular tachycardia, complete heart block, and ventricular tachycardia 21 .

Puntmann et al in a study of 100 patients, who recovered from COVID-19, described cardiac involvement in 78 % of patients and ongoing myocardial inflammation in 60 % of patients by cardiac magnetic resonance (CMR) independents of preexisting comorbidities, the severity of the acute illness, and time from the original diagnosis 19 . Native T1 and T2 showed the best discriminatory power to detect COVID-19 related myocardial damage (AUC of 0,83 and 0,82) 19 .

Another CMR imaging study among 26 competitive college athletes with SARS-COV 2 infection revealed that 46% of enrolled patients had evidence of myocarditis or prior myocardial injury 22 

Recently, elevated D-Dimer and C reactive protein levels have been reported in 30% and 9,5 % of survivors, respectively 24 . A retrospective observational cohort study of 163 patients with confirmed COVID-19 not receiving anticoagulation reported an incidence of thrombosis, including arterial and venous events, of 2,5 % (95% CI, 0,8% to 7,6%) at day 30 after discharge. The same study revealed hemorrhagic events during the follow-up period (6 of 163; 3,7%) 25 . Despite recovery from COVID-19 infection, survivors may still live thrombosis due to endothelial dysfunction with a high prevalence of pulmonary vascular dysfunction 16 .

In view of this emerging signal, the British Thoracic Society suggests a computed tomography angiography (CTA) in patients with suspected pulmonary embolism 16 . Another follow-up algorithm for the investigation of patients after COVID-19 is proposed by Dwahan et al. with perfusion imaging as a key tool in the triage tree, including VQ scintigraphy and dual-energy CT 26 .

Also, high-performance low field MRI may provide a morphology and function assessment in a single examination without the need to apply intravenous contrast agents or a radiation exposure 27 . survivors, reporting brain fog (81%), headache (68%), numbness (60%), dysgeusia (59%), anosmia (55%), and myalgias (55%) 31 . See figure 5 . In a prospective study of 6-months outcomes of COVID-19 patients, abnormalities in cognitive and functional outcomes were reported in >90 % of survivors. The outcomes were assessed with different scales, such as the modified Ranking scale, Barthel index (for activities of daily living), and Quality of Life in Neurological Disorder ( a measurement of anxiety, depression, fatigue, and sleep disorders).In a multivariate analysis, persistent neurological manifestations were independent predictors of impaired activities of daily living and were inversely associated with return to work 32 .

COVID-19 survivors have reported a spectrum of persistent or presenting psychiatric symptoms after initial infection, with a high prevalence in the initial phase after discharge 33 

A recent prospective study on brain MRI findings indicated that COVID-19 survivors with no specific neurological symptoms exhibited brain microstructure abnormalities and a decrease in cerebral blood flow after a 3-months follow-up 36 

Various guidelines treated long-Covid-19 management or have included a recommendation for long-COVID in their guidelines for COVID-19, NICE guidelines suggested a multidisciplinary approach to identify, refer, and treat these patients 38 . In this scenario, also imaging should play a crucial role. Nuzzo et al. proposed an MRI control of the brain in the management of long-COVID-19 patients with suspected neurological sequelae, in order to assess the potentially neuronal degeneration due to microvascular disorders 39, 40 .

Various abdominal manifestations were reported in COVID-19 survivors, including renal, digestive, and metabolic sequelae. In a cohort of 287 survivors from COVID-19, approximately 1,4 % of patients experienced renal failure in long-term follow-up 41 . At 6 months after acute infection, 35%

of COVID-19 survivors demonstrated a decreased estimated glomerular filtration rate 13 . More importantly, the study reported that 13 % of patients developed a new-onset reduction of estimated glomerular filtration rate after acute infection 13 .

The prospective COVERSCAN study evaluated the medium-term organ impact of the long-COVID syndrome in different organs, including kidneys (4%), liver (28%), pancreas (40%), and spleen (4%), with single-organ and multiorgan impairment in 70% and 29%, respectively 4 

In the era of modern medicine, artificial intelligence (AI) is a growing field of interest in diagnostic imaging due to technological innovations that have led to constant development 46, 47 .

During the pandemic, the number of imaging investigations performed has been increasing dramatically and it will continue to grow due to the persistent symptomatology in Long-COVID patients 48 . AI has numerous potential applications in diagnostic imaging, including image analysis, decision-making, and prognosis prediction 49 (Figure 6) , and has been widely used in the fight against the COVID-19 pandemic 50-53 .

For example, AI has been shown to be able to differentiate COVID-19 pneumonia from community-acquired pneumonia and other lung conditions 54, 55 . A deep learning model achieved high sensitivity (90%) and high specificity (96%)in the detection of COVID-19 using chest CT with an area under the curve of 0.96 and an average time for each CT scan of 4,51 s 54 . In addition, an AI algorithm can identify patients who developed severe COVID-19 symptoms 56, 57 . Quiroz-Juarez has presented a study for the early identification of high-risk patients among those exposed to the SARS-COV-2 virus, using a supervised artificial neural network 56 . The machine learning models were trained using comorbidities, patient demographic data, as well as recent COVID-19-related medical information. The authors reported that the disease outcome can be predicted with specificity greater than 82%, sensitivity greater than 86%, and accuracy greater than 84% 56 .

After the COVID-19 pandemic, an increasing number of survivors continue to battle the symptoms of the disease. Various studies have found that a fraction of COVID-19 survivors developed fibrotic abnormalities 12, 58 . Zou et al. evaluated an AI-assisted chest CT technology to quantitively measure the extent and the degree of pulmonary inflammation in 239 patients that developed pulmonary fibrosis after COVID-19 pneumonia at 30, 60, and 90 days after discharge 59 . The authors reported that the AI inflammation score showed a good correlation with the quantitative pulmonary fibrosis score, concluding that AI-assisted chest CT technology may provide qualitative and quantitative data to analyze the long-term evolution of pulmonary fibrosis 59 .

In addition, AI has already proven to be an excellent tool in the diagnosis and stratification of patients with pulmonary fibrosis 60 

It is well known that SARS-COV-2 involved microcirculation, leading to endothelial cell and pericytes damage, micro-thrombosis, and capillary congestion that may contribute to persisting COVID-19 symptoms 63 . In addition, the pro-inflammatory state caused by SARS-COV-2 is considered a trigger in promoting plaque vulnerability 64, 65 Mohamud et al. described a case series of 6 patients with COVID-19 presenting to the emergency department with acute ischemic infarction due to intraluminal carotid artery thrombus 65 . Similar findings have also been described for coronary artery disease 66 , given the relationship between carotid and coronary atherosclerosis 67 . The application of AI algorithms in cardiovascular imaging is an evolving field that has shown to be an efficient tool for the diagnosis of atherosclerotic disease as well as for plaque characterization 46 . Various AI-trained models that combined imaging data and clinical risk factors have been applied to aid in predicting outcomes [68] [69] [70] [71] 72 . AI models can also help to predict the structure of protein essential for the replication of the virus.

A recent study published in Nature in 2020 evaluated a deep-learning approach to protein structure prediction with the purpose to understand the fold shape, the detailed side-chain configurations in binding regions as well as to target mutations to destabilize the protein, improving the biological insight 78 . DeepTracer, a fully automated software based on a customized deep convolutional neural network, was used to derive macromolecules' 3D maps at a nearatomic resolution of SARS-COV-2 from electron cryomicroscopy 79 An emerging issue is the rapid spread of COVID-19 variants with serious concerns to global health, it is important to be able to promptly identify and detect these emerging variants. Perez-Romero et al. proposed a deep-learning algorithm that was able to deliver the primer sets for each variant, reporting an accuracy above 95 %, with two exceptions, in particular for the variants P1-2 with an accuracy of 88,99 % and for B.1.1.519 with 60,40% % for the forward primer, and 64,32% for the reverse primer 83 . 

The multi-organ involvement of COVID-19 beyond the acute infection is now recognized in a fraction of the patients, namely "long-COVID". Given the high number of patients affected by COVID-19 and many patients with symptoms suggestive for long-COVID, early identification and management are vital to improving patient outcomes. In the near future, the adoption of AI may greatly enrich daily clinical practice with the promise to provide solutions to the physician interrogations of risk stratification and the clinical progression of COVID-19 patients. 3a) showing bilateral interlobular thickening with scattered reticular and ground glass infiltration. Follow-up HRCT was done 6 months from start of symptoms revealing a worsening with bilateral fibrotic-like changes, represents by parenchimal bands and peri-bronchial thickening (fig 3b) . Dear Editor,

We would like to propose to your attention for potential publication the manuscript entitled "Long-COVID diagnosis: from diagnostic to advanced AI-driven models".

Most of the patients infected with SARS-COV2 fully recovered within some weeks. In a considerable number of patients, like many other viral infections, various long-lasting symptoms have been described. Although there is a scarcity of literature on this topic, it is not infrequent to have patients presenting with long-term symptoms, now defined as "long COVID-19 syndrome". Given the high number of contagious over the world, it is necessary to understand and comprehend this emerging pathology to enable early diagnosis and improve patents outcomes. In this scenario, AI-based models can be applied in long-COVID-19 patients to assist clinicians and at the same time, to reduce the considerable impact on care and rehabilitation unit. The purpose of this manuscript is to review different aspects of long-COVID-19 syndrome from clinical presentation to diagnosis, highlighting the considerable impact that AI can have Thank you for your time and consideration, On behalf of all Authors, Sincerely,

CT findings of COVID-19 pneumonia in ICU-patients

Complications in COVID-19 patients: Characteristics of pulmonary embolism

Imaging in COVID-19-related myocardial injury

Multiorgan impairment in low-risk individuals with post-COVID-19 syndrome: a prospective, community-based study

Post-COVID-19 syndrome: epidemiology, diagnostic criteria and pathogenic mechanisms involved

More than 50 long-term effects of COVID-19: a systematic review and meta-analysis

Group GAC-19 P-ACS. Persistent Symptoms in Patients After Acute COVID-19

NICE guideline on long COVID

Post-COVID syndrome in non-hospitalised patients with COVID-19: a longitudinal prospective cohort study

Impact of severe acute respiratory syndrome (SARS) on pulmonary function, functional capacity and quality of life in a cohort of survivors

Follow-up chest radiographic findings in patients with MERS-CoV after recovery

Cardiopulmonary recovery after COVID-19: an observational prospective multicentre trial

6-month consequences of COVID-19 in patients discharged from hospital: a cohort study

Persistent Post-COVID-19 Interstitial Lung Disease. An Observational Study of Corticosteroid Treatment

COVID -19 pneumonia imaging follow -up : when and how ? A proposition from ESTI and ESR European Society of Radiology

Respiratory follow-up of patients with COVID-19 pneumonia

NICE guideline on long COVID

Long COVID-19: A Primer for Cardiovascular Health Professionals, on Behalf of the CCS Rapid Response Team

Outcomes of Cardiovascular Magnetic Resonance Imaging in Patients Recently Recovered From Coronavirus Disease 2019 (COVID-19)

Incidence of Stress Cardiomyopathy During the Coronavirus Disease

Management of Arrhythmias Associated with COVID-19

Cardiovascular Magnetic Resonance Findings in Competitive Athletes Recovering From COVID-19 Infection

Cardiovascular magnetic resonance evaluation of soldiers after recovery from symptomatic SARS-CoV-2 infection: a case-control study of cardiovascular post-acute sequelae of SARS-CoV-2 infection (CV PASC)

Long-COVID': a cross-sectional study of persisting symptoms, biomarker and imaging abnormalities following hospitalisation for COVID-19

Postdischarge thrombosis and hemorrhage in patients with COVID-19

Beyond the clot: perfusion imaging of the pulmonary vasculature after COVID-19

High-performance low field MRI enables visualization of persistent pulmonary damage after COVID-19

Long COVID: understanding the neurological effects

Potential Neurologic Manifestations of COVID-19

Post-acute COVID-19 syndrome

Persistent neurologic symptoms and cognitive dysfunction in non-hospitalized Covid-19 "long haulers

A prospective study of long-term outcomes among hospitalized COVID-19 patients with and without neurological complications

Psychological Distress and Its Correlates Among COVID-19 Survivors During Early Convalescence Across Age Groups

Anxiety and depression in COVID-19 survivors: Role of inflammatory and clinical predictors

6-month neurological and psychiatric outcomes in 236 379 survivors of COVID-19: a retrospective cohort study using electronic health records. The lancet Psychiatry

Long-term microstructure and cerebral blood flow changes in patients recovered from COVID-19 without neurological manifestations

Persistent white matter changes in recovered COVID-19 patients at the 1-year follow-up

COVID-19 rapid guideline: managing the long-term effects of COVID-19 NICE guideline; c2020

Long-Term Brain Disorders in Post Covid-19 Neurological Syndrome (PCNS) Patient

Microvascular Injury in the Brains of Patients with Covid-19

Assessment and characterisation of post-COVID-19 manifestations

Liver function recovery of COVID-19 patients after discharge, a follow-up study

Newly diagnosed diabetes mellitus, DKA, and COVID-19: Causality or coincidence? A report of three cases

New-onset diabetes in "long COVID

Risk of clinical sequelae after the acute phase of SARS-CoV-2 infection: retrospective cohort study

Artificial Intelligence in Computed Tomography Plaque Characterization: A Review

Potential Role of Artificial Intelligence in Cardiac Magnetic Resonance Imaging

Effect of COVID-19 on computed tomography usage and critical test results in the emergency department: an observational study. C open

Artificial intelligence in pulmonary medicine: computer vision, predictive model and COVID-19

COVLIAS 1.0: Lung Segmentation in COVID-19 Computed Tomography Scans Using Hybrid Deep Learning Artificial Intelligence Models

The role of artificial intelligence in tackling COVID-19. Future Virol

53. van Ginneken B. The Potential of Artificial Intelligence to Analyze Chest Radiographs for Signs of COVID-19 Pneumonia

Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy

Artificial intelligence for the detection of COVID-19 pneumonia on chest CT using multinational datasets

Identification of high-risk COVID-19 patients using machine learning

Predicting the Disease Outcome in COVID-19 Positive Patients Through Machine Learning: A Retrospective Cohort Study With Brazilian Data

Post covid 19 pulmonary fibrosis. Is it real threat?

The characteristics and evolution of pulmonary fibrosis in COVID-19 patients as assessed by AI-assisted chest HRCT

Computer-Aided Diagnosis of Pulmonary Fibrosis Using Deep Learning and CT Images

Artificial intelligence identifies inflammation and confirms fibroblast foci as prognostic tissue biomarkers in idiopathic pulmonary fibrosis

Artificial intelligence for prediction of COVID-19 progression using CT imaging and clinical data

SARS CoV-2 related microvascular damage and symptoms during and after COVID-19: Consequences of capillary transit-time changes, tissue hypoxia and inflammation

Inflammation, plaque progression and vulnerability: evidence from intravascular ultrasound imaging

Intraluminal Carotid Artery Thrombus in COVID-19: Another Danger of Cytokine Storm?

Possible mechanisms responsible for acute coronary events in COVID-19

Insight from imaging on plaque vulnerability: similarities and differences between coronary and carotid arteries-implications for systemic therapies

Machine learning for prediction of all-cause mortality in patients with suspected coronary artery disease: a 5-year multicentre prospective registry analysis

Machine learning of clinical variables and coronary artery calcium scoring for the prediction of obstructive coronary artery disease on coronary computed tomography angiography: analysis from the CONFIRM registry

Maximization of the usage of coronary CTA derived plaque information using a machine learning based algorithm to improve risk stratification; insights from the CONFIRM registry

Novel Application of Artificial Intelligence Algorithms to Develop a Predictive Model for Major Adverse Neurologic Events in Patients With Carotid Atherosclerosis

Improved long-term prognostic value of coronary CT angiography-derived plaque measures and clinical parameters on adverse cardiac outcome using machine learning

Stacking Ensemble-Based Intelligent Machine Learning Model for Predicting Post-COVID-19 Complications

Deep learning for pulmonary embolism detection on computed tomography pulmonary angiogram: a systematic review and meta-analysis

Three-month follow-up of pulmonary embolism in patients with COVID-19

Predicting COVID-19 in China Using Hybrid AI Model

Forecasting the long-term trend of COVID-19 epidemic using a dynamic model

Improved protein structure prediction using potentials from deep learning

DeepTracer for fast de novo cryo-EM protein structure modeling and special studies on CoV-related complexes

Genomic mutations and changes in protein secondary structure and solvent accessibility of SARS-CoV-2 (COVID-19 virus)

Viral reverse engineering using Artificial Intelligence and big data COVID-19 infection with Long Short-term Memory (LSTM)

AI-guided discovery of the invariant host response to viral pandemics

Design of Specific Primer Sets for the Detection of SARS-CoV-2 Variants of Concern B.1.1.7, B.1.351, P.1, B.1.617.2 using Artificial Intelligence. bioRxiv

All authors agreed with the content and gave consent to submit.All authors contributed equally as authors to this work.The authors state that this work is not under consideration elsewhere and none of the paper's contents have been previously published.The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.All authors read and approved the final manuscript.The scientific guarantor of this publication is the corresponding authorThe authors declare that they have no competing interests.All authors agreed with the content and gave consent to submit.All authors contributed equally as authors to this work.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-forprofit sectors.All authors read and approved the final manuscript.The scientific guarantor of this publication is the corresponding authorThe authors declare that they have no competing interests.