Artificial intelligence and pneumonia: a rapidly evolving frontier

Abstract

The use of artificial intelligence is increasing and permeates many aspects of our daily lives. Its application in medicine is also increasing, supported by the increased digitalisation of clinical data. The COVID-19 pandemic accelerated innovation in artificial-intelligence-assisted medical care, particularly around lung imaging and respiratory sounds, which have applications for pneumonia.1Jia LL Zhao JX Pan NN et al.Artificial intelligence model on chest imaging to diagnose COVID-19 and other pneumonias: a systematic review and meta-analysis.Eur J Radiol Open. 2022; 9100438Summary Full Text Full Text PDF PubMed Scopus (6) Google Scholar Machine-leaning technologies and algorithms that enable the identification of patterns have been used with chest radiography, CT, ultrasound, and MRI as well as with cough and lung sounds to aid in the diagnosis of pneumonia and other respiratory diseases and in clinical decision-making. Using a complex structure of algorithms based on multilayered convolutional neural networks with large numbers of parameters that are trained on massive amounts of data, deep learning automates high-level feature extraction to produce accurate insights and predictions. The diagnostic performance of deep-learning models based on medical imaging for many conditions and diseases has been reported to be equivalent or superior to that of health-care professionals.2Liu X Faes L Kale AU et al.A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis.Lancet Digit Health. 2019; 1: e271-e297Summary Full Text Full Text PDF PubMed Scopus (730) Google Scholar Pneumonia is typically diagnosed on the basis of non-specific symptoms and physical examination findings, which can lead to missed or incorrect diagnoses, treatments, complications, and deaths. As a valuable source of diagnostic and prognostic information for pneumonia and COVID-19, medical imaging is subject to human operation and interpretation as well as human resource and training challenges. Artificial-intelligence-assisted automated and accurate diagnostics and prognostics that provide real-time analysis, interpretation, and clinical decision-making support have the potential to address these challenges. For example, an externally validated deep-learning algorithm for triaging patients at fever clinics in China with suspected COVID-19 based on chest CT showed high accuracy for diagnosing COVID-19 across populations with varied COVID-19 prevalence.3Wang M Xia C Huang L et al.Deep learning-based triage and analysis of lesion burden for COVID-19: a retrospective study with external validation.Lancet Digit Health. 2020; 2: e506-e515Summary Full Text Full Text PDF PubMed Scopus (59) Google Scholar In another study of patients with COVID-19 in the USA, an externally validated artificial-intelligence-assisted chest radiography model had better prognostic performance for predicting progression to critical illness than did clinical data or radiologist-derived severity scores.4Jiao Z Choi JW Halsey K et al.Prognostication of patients with COVID-19 using artificial intelligence based on chest x-rays and clinical data: a retrospective study.Lancet Digit Health. 2021; 3: e286-e294Summary Full Text Full Text PDF PubMed Scopus (0) Google Scholar If both of these artificial-intelligence-assisted technologies can be extended to pneumonia, and other respiratory illnesses, they have the potential to help relieve human resource constraints and to improve efficiency, clinical workflow, and outcomes where these imaging modalities are available. Deep-learning models are also being used to develop cough analysis and digital stethoscope technologies to aid in the assessment of pneumonia and other respiratory illnesses.5Sethi AK Muddaloor P Anvekar P et al.Digital pulmonology practice with phonopulmography leveraging artificial intelligence: future perspectives using dual microwave acoustic sensing and imaging.Sensors (Basel). 2023; 235514Crossref Scopus (0) Google Scholar, 6Sfayyih AH Sabry AH Jameel SM et al.Acoustic-based deep learning architectures for lung disease diagnosis: a comprehensive overview.Diagnostics (Basel). 2023; 131748PubMed Google Scholar, 7Sharan RV Rahimi-Ardabili H Detecting acute respiratory diseases in the pediatric population using cough sound features and machine learning: a systematic review.Int J Med Inform. 2023; 176105093Crossref PubMed Scopus (0) Google Scholar Globally, a cough is the most common patient-reported reason for seeking health care. Deep-learning algorithms are being used for the recognition of coughs, lung sounds, and breathing patterns.7Sharan RV Rahimi-Ardabili H Detecting acute respiratory diseases in the pediatric population using cough sound features and machine learning: a systematic review.Int J Med Inform. 2023; 176105093Crossref PubMed Scopus (0) Google Scholar, 8Zhang J Wang HS Zhou HY et al.Real-world verification of artificial intelligence algorithm-assisted auscultation of breath sounds in children.Front Pediatr. 2021; 9627337Google Scholar Often subjective, variable, and open to interpretation by the clinician, objective and consistent cough and lung sound evaluations facilitated by artificial intelligence could greatly enhance clinical assessments of respiratory diseases, including pneumonia. Given that pneumonia affects the most vulnerable populations, artificial-intelligence-assisted technologies have the potential to transform health care, improving the accessibility, speed, accuracy, reliability, generalisability, quality, effectiveness, and delivery of health-care services as well as health outcomes. Artificial-intelligence-assisted imaging and acoustic technologies might be more effective than human senses, particularly with smaller or more subtle visual differences or with lower or higher frequency sounds. In addition to expanding access to services in resource-constrained, rural, or remote settings without highly trained health-care providers with specialised imaging expertise or lung auscultation experience, these artificial-intelligence-assisted technologies might have substantial cost savings.9Khanna NN Maindarkar MA Viswanathan V et al.Economics of artificial intelligence in healthcare: diagnosis vs treatment.Healthcare (Basel). 2022; 102493PubMed Google Scholar The benefits of artificial intelligence also include improving human performance, democratising medical knowledge and excellence, automating drudgery, and allocating limited resources.10Price WN Risks and remedies for artificial intelligence in health care.https://www.brookings.edu/articles/risks-and-remedies-for-artificial-intelligence-in-health-care/Date: Nov 14, 2019Date accessed: September 23, 2023Google Scholar However, several limitations and challenges need to be overcome for artificial intelligence to become a powerful tool in the diagnosis and management of patients with pneumonia. These limitations include the subjectiveness in reference standards (ground truths), inadequate volume of data or low-quality data, insufficient training data and evaluation in real-world clinical settings, the need to externally validate models, lack of transparency and adequate reporting of diagnostic accuracy, and regulatory challenges.2Liu X Faes L Kale AU et al.A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis.Lancet Digit Health. 2019; 1: e271-e297Summary Full Text Full Text PDF PubMed Scopus (730) Google Scholar, 10Price WN Risks and remedies for artificial intelligence in health care.https://www.brookings.edu/articles/risks-and-remedies-for-artificial-intelligence-in-health-care/Date: Nov 14, 2019Date accessed: September 23, 2023Google Scholar Furthermore, existing or emerging risks such as artificial intelligence system errors, biases, and data privacy and security breaches that could cause patient harm or increase health inequities need to be anticipated and addressed. Although resource-constrained settings have the potential to benefit from artificial intelligence the most, they are also the most vulnerable to potential harm. These artificial-intelligence-assisted technologies are tools and do not substitute for the strengthening of health systems. As we commemorate yet another World Pneumonia Day in a world where pneumonia, stubbornly, remains the most deadly communicable disease and artificial intelligence is a rapidly evolving frontier, we recognise the persistent unmet need for improved pneumonia diagnostics, prognostics, and treatment options, and encourage continued innovation, research, and evidence-based approaches for pneumonia prevention and management while also taking care to ensure responsible, sustainable, and inclusive development and use of artificial intelligence-assisted pneumonia-related technologies. ASG is a paid consultant for Caption Health. EDM holds grants from the Bill & Melinda Gates Foundation, US National Institutes of Health, US Agency for International Development, US Centers for Disease Control and Prevention, and Thrasher Research Fund for studies in lower respiratory infection; receives funding from Moderna for respiratory syncytial virus prevention; and is a paid scientific consultant to Sonavi Labs. This funding arrangement has been reviewed and approved by the Johns Hopkins University in accordance with its conflict of interest policies.

Ähnliche Arbeiten