Diagnosis of idiopathic pulmonary fibrosis using artificial intelligence:a systematic review and meta-analysis

Background and aims: Idiopathic pulmonary fibrosis (IPF) is a fatal fibrosing interstitial lungdisease with challenging Diagnosis and varying disease progression. Need for expensive andinvasive procedures for diagnostic confirmation, lack of effective early diagnostic tools and nonspecificpresentation contribute to this challenge. It causes irreversible damage to lung tissue butearly and proper diagnosis can help to control this disease. Lately, artificial intelligence (AI) hasapproached significant advances in medical image analysis. We aimed to systematically evaluatethe application of AI for diagnosis of this disease.Method: Systematic search in PubMed, Scopus, Embase and Web of Science was conducted upto March ۲۰۲۳. Appropriate terms including Idiopathic pulmonary fibrosis, artificial intelligenceand other relevant terms were searched. All articles were independently reviewed by two reviewersusing Rayyan software. Abstracts and article helpfulness were assessed and duplicates wereeliminated. Then, inclusion and exclusion criteria were applied. Animal studies, books and bookchapters were excluded. Quality assessment was performed with the revised Quality Assessmentof Diagnostic Accuracy Studies (QUADAS-۲) tool. Microsoft Excel was used for data extraction.Meta-analysis was conducted using CMA ۳.۷ software.Results: Out of ۱۱۱۲ studies, five studies were found to be eligible for this study. Imaging type,sample size, machine learning algorithms, accuracy, sensitivity, specificity and area under theROC curve (AUC) were evaluated. The Ultimate AUC of ۰.۸۴۵ was determined (AUC = ۰.۸۴۵,۹۵% CI = low ۰.۸۴۱-high ۰.۸۴۹, p-value < ۰.۰۰۱).Conclusion: Studies showed acceptable performance using numerous recognized deep learningmodels in the task of IPF diagnosis. In some studies, models demonstrated better accuracy thanradiologists. Overall studies showed that automated diagnostic tools can serve as an advantageousclinical aid for diagnosing the disease. Also, there is demand for multidisciplinary diagnosis inidiopathic pulmonary fibrosis and artificial intelligence methods cannot fully replace it. Morestudies should be done to explore the development of IPF at baseline and follow-up, in additionto assess the efficacy of anti-fibrotic treatment.

کلیدواژه ها:

Artificial Intelligence ، Idiopathic Pulmonary Fibrosis ، Interstitial Lung Disease ، Diagnosis