Skip to main content
Download PDF

Diagnosis accuracy of machine learning for idiopathic pulmonary fibrosis: a systematic review and meta-analysis

European Journal of Medical Research volume 30, Article number: 288 (2025) Cite this article

Abstract

Background

The diagnosis of idiopathic pulmonary fibrosis (IPF) is complex, which requires lung biopsy, if necessary, and multidisciplinary discussions with specialists. Clinical diagnosis of the two ailments is particularly challenging due to the impact of interobserver variability. Several studies have endeavored to utilize image-based machine learning to diagnose IPF and its subtype of usual interstitial pneumonia (UIP). However, the diagnostic accuracy of this approach lacks evidence-based support.

Objective

We conducted a systematic review and meta-analysis to explore the diagnostic efficiency of image-based machine learning (ML) for IPF.

Data sources and methods

We comprehensively searched PubMed, Cochrane, Embase, and Web of Science databases up to August 24, 2024. During the meta-analysis, we carried out subgroup analyses by imaging source (computed radiography/computed tomography) and modeling type (deep learning/other) to evaluate its diagnostic performance for IPF.

Results

The meta-analysis findings indicated that in the diagnosis of IPF, the C-index, sensitivity, and specificity of ML were 0.93 (95% CI 0.89–0.97), 0.79 (95% CI 0.73–0.83), and 0.84 (95% CI 0.79–0.88), respectively. The sensitivity of radiologists/clinicians in diagnosing IPF was 0.69 (95% CI 0.56–0.79), with a specificity of 0.93 (95% CI 0.74–0.98). For UIP diagnosis, the C-index of ML was 0.91 (95% CI 0.87–0.94), with a sensitivity of 0.92 (95% CI 0.80–0.97) and a specificity of 0.92 (95%CI 0.82–0.97). In contrast, the sensitivity of radiologists/clinicians in diagnosing UIP was 0.69 (95% CI 0.50–0.84), with a specificity of 0.90 (95% CI 0.82–0.94).

Conclusions

Image-based machine learning techniques demonstrate robust data processing and recognition capabilities, providing strong support for accurate diagnosis of idiopathic pulmonary fibrosis and usual interstitial pneumonia. Future multicenter large-scale studies are warranted to develop more intelligent evaluation tools to further enhance clinical diagnostic efficiency.

Trial registration This study protocol was registered with PROSPERO (CRD42022383162).

Introduction

Idiopathic pulmonary fibrosis (IPF) is a chronic interstitial lung disease of unknown origin, marked by progressive lung constriction. A study published in 2021 indicated that the incidence rate ranged from 0.9 to 13.0 per 100,000 people, and the prevalence rate varied from 3.3 to 45.1 per 100,000 people [1]. IPF, as the most prevalent idiopathic interstitial pneumonia (IIP), is linked to the poorest prognosis, and the estimated median survival period post-diagnosis is 3–5 years [2,3,4,5]. Early diagnosis is imperative for tailoring treatment plans, such as choosing between anti-fibrotic treatment and the treatment of pulmonary fibrosis due to other causes [6]. The diagnosis of IPF necessitates multidisciplinary discussions and collaborations among clinicians, radiologists, and pathologists [7]. However, the prolonged time from referral to multidisciplinary diagnosis is notable due to interobserver variabilities [8, 9]. Therefore, a diagnostic method that mitigates observer variability, and can accurately and swiftly differentiate between IPF and non-IPF interstitial lung diseases, is necessary in clinical practice.

The Diagnosis and Treatment Guidelines for Idiopathic Pulmonary Fibrosis for the first time included radiological UIP patterns in the definition of IPF, emphasizing the importance and diagnostic role of identifying high-resolution computed tomography (HRCT) UIP patterns. With the increasing importance of medical imaging in precision medicine for disease diagnosis, prognosis, and treatment planning [10], computed tomography (CT) emerges as a valuable tool providing visual data to enhance decision-making [11]. However, qualitative CT assessment remains challenging, leading to common discrepancies even among experienced experts [12, 13]. Hence, there is an urgent demand for an automated clinical tool that can assist clinicians in making precise and timely diagnoses.

Computer-aided diagnosis empowers doctors to leverage information technology (IT) to interpret and utilize various imaging techniques. The primary objective is to shorten diagnosis time and enhance accuracy, with IT serving as a supportive or even independent diagnostic option [14]. Computer-aided diagnostic algorithms fall within the realm of artificial intelligence (AI), mimicking human thinking. With the increasing wealth of imaging data and the availability of computing resources, AI is gaining popularity [15]. Quantitative imaging techniques in medical imaging are increasingly used in an exponential way [16]. In this context, some researchers are striving to develop tools that can assist in the diagnosis of IPF and UIP. Nevertheless, there is presently insufficient evidence-based backing for the detailed diagnostic value. To address this gap, we conducted a systematic review and meta-analysis to explore the diagnostic efficacy of image-based machine learning (ML) for IPF and UIP.

Methods

Study registration

Our study adhered to the preferred reporting guidelines (PRISMA 2020) for systematic review and was prospectively registered on Prospero (CRD42022383162).

Eligibility criteria

Inclusion criteria

  1. (1)

    The subjects were patients with suspected interstitial lung disease.

  2. (2)

    The types of studies included case–control study, cohort study, case–control study, and case-cohort study.

  3. (3)

    A comprehensive ML model was developed to identify the prognosis of IFP or UPF or interstitial lung disease.

  4. (4)

    Studies lacking external validation are also incorporated.

  5. (5)

    Various ML studies published in the same dataset.

  6. (6)

    Studies reported in English.

Exclusion criteria

  1. (1)

    Research types: Meta-analysis, review, guidelines, expert opinions, and conference abstracts published without peer review.

  2. (2)

    Only an analysis of risk factors was conducted, and a comprehensive ML model was not constructed.

  3. (3)

    The following outcome indexes to assess the accuracy of ML model are missing.

  4. (4)

    Validation of the maturity scale only.

Data sources and search strategy

On August 24, 2024, we systematically searched relevant literature in Cochrane, Embase, and Web of Science databases using a combination of subject terms and free words (with the start time being the establishment time of each database). Detailed information regarding the search materials is available in Table S1.

Study selection and data extraction

We imported the identified literature into Endnote20.0 and, following the removal of duplicates, checked the titles and abstracts. Subsequently, we downloaded the full texts, thoroughly read them, and selected documents that aligned with the objectives of our study. Before data extraction, we established a standardized data extraction table to capture essential information such as author details, publication year, study type, diagnostic events, training set, validation set, type of model used, C-index, sensitivity, and specificity. The above literature screening and data extraction were carried out by two researchers (LC & YC) independently, and cross-checking was carried out after completion. In case of any dispute, the involvement of the third researcher (LPC) was consulted to assist in the decision.

Risk of bias in studies

We utilized QUADAS-2 to evaluate the risk of bias in the original studies. This evaluation comprised numerous questions distributed across four distinct areas: participants, predictive variables, results, and statistical analysis. These areas encompassed 2, 3, 6, and 9 specific questions, respectively, each with three response options (yes/possibly yes, no/probably no, and no information). A domain was deemed to be at high risk if at least one question was answered by no or probably no. Conversely, to be classified as low risk, all questions in a domain should be answered by yes or possibly yes. The overall risk of bias was considered low when all areas were deemed low risk and high when at least one domain was categorized as high risk.

The two researchers (LC & XLH) independently conducted the bias risk assessment using QUADAS-2 and cross-checked their results after completion. In case of any dispute, the third researcher (CW) was consulted to assist in the adjudication.

Outcomes

The C-index was used as the primary outcome to depict the prediction precision of the ML models. Meanwhile, the sample size of the case group and the control group was seriously unbalanced, which was insufficient to reflect the prediction accuracy of the case group when the difference was enlarged. Therefore, the sensitivity and specificity of ML are also used as outcome measures.

Synthesis methods

We performed a meta-analysis to evaluate the overall accuracy indicator (c-index) for assessing ML models. In instances where the original studies did not provide a 95% confidence interval and standard error for the c-index, we consulted the research of Debray TP et al. [17] to estimate its standard error. Given the variations in the included variables and the inconsistency of parameters across different ML models, the random effects model was prioritized for the meta-analysis of the c-index.

Additionally, we conducted a meta-analysis of sensitivity and specificity employing a bivariate mixed-effects model. During the meta-analysis process, sensitivity and specificity were evaluated based on the diagnostic fourfold table. However, a majority of original studies did not report the diagnostic fourfold table. In such instances, we utilized two approaches to calculate the diagnostic fourfold table: 1. The diagnostic table was computed based on sensitivity, specificity, precision, and the number of cases; 2. Extraction of sensitivity and specificity were extracted based on the optimal Youden’s index, followed by calculation with the number of cases. The meta-analysis for this study was conducted using R 4.2.0 (R development Core Team, Vienna, http://www.R-project.org).

Results

Study selection

A total of 3572 articles were retrieved in the initial search. Following the elimination of duplicates and irrelevant articles, 74 articles underwent a detailed full-text reading. Among them, we excluded unpublished conference abstracts, studies that only performed image segmentation and reconstruction, and studies that did not assess the diagnostic performance of outcome indicators for the disease. Ultimately, 16 articles [18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33] were included in this study, as illustrated in Fig. 1.

Fig. 1
figure 1

The literature screening process

Full size image

Study characteristics

The 16 included studies were published from 2004 to 2024, encompassing a total of 7209 study subjects, predominantly from North America, Europe, and Japan. Among these, 6 articles [18,19,20,21,22] focused on the diagnosis of IPF (Table 1). Notably, one of these articles [21] employed chest X-ray as a modeling variable, while the remaining four utilized high-resolution CT. The remaining 10 articles [23,24,25,26,27,28] centralized around the diagnosis of UIP. Within this subset, eight articles [23, 25,26,27] leveraged ML for the diagnosis of UIP patterns under HRCT, and two articles [24, 28] predicted pathological UIP patterns, all employing high-resolution CT as modeling variables. Across the 11 pieces of literature, ML algorithms predominantly included artificial neural network (ANN), convolutional neural network (CNN), and vector space model (SVM). Notably, seven studies [20,21,22,23,24,25,26] compared the diagnostic performance between human radiologists/clinical experts and ML models, as illustrated in Fig. 2.

Table 1 Basic Information on the Included Literature
Full size table
Fig. 2
figure 2

(A) Risk of Bias Summary for Included Primary Studies, and (B) Risk of Bias Graph for Included Primary Studies

Full size image

Risk of bias in studies

All included studies adopted a case–control design, and only one study adopted a non-deep learning method to construct a model. In the construction of conventional ML, non-deep learning modeling variables needed to be manually encoded, which may introduce a higher bias risk, especially in the context of case–control studies. The impact of other image-based deep learning on case–control studies was relatively minimal. Consequently, there is a heightened risk of bias in the selection of cases. Various ML methods in the primary studies employed their respective evaluation criteria for the interpretation of the gold standard. This variability in evaluation criteria may introduce bias. However, the impact is considered low when implementing or interpreting the index tests, and we believe there is no high bias risk (Tables 2 and 3). Detailed assessment results are illustrated in Fig. 2.

Table 2 Quality evaluation of the trials
Full size table
Table 3 Risk of bias assessment results
Full size table

Meta-analysis

IPF

ML model

In the diagnosis of IPF, the combined C-index, sensitivity, and specificity of ML were 0.93 (95% CI 0.89–0.97), 0.79 (95% CI 0.73–0.83), and 0.84 (95% CI 0.79–0.88), respectively. Notably, the model with the highest diagnostic performance was the deep learning model (CT) proposed by Wenxi Yu [21], with a C-index of 0.99 (95% CI 0.97–1), as illustrated in Figs. 3 and 4.

Fig. 3
figure 3

Forest Map of Meta-Analysis of C-index for ML in Diagnosis of IPF

Full size image
Fig. 4
figure 4

Forest Map of Meta-analysis of the Sensitivity and Specificity of ML in Diagnosis of IPF

Full size image
Radiologists/clinicians

Among the six included pieces of literature on IPF [18,19,20,21,22], three of them [20,21,22] compared ML with human experts. In these studies, the sensitivity of radiologists/clinicians in the diagnosis of IPF was notably lower compared to that of ML. However, it was essential to note that due to the small sample size, only observational analyses were performed. In these three pieces of literature, the sensitivity of radiologists/clinicians diagnosing without the assistance of ML models ranged from 57 to 85%, while the specificity ranged from 75 to 98% (Fig. 5).

Fig. 5
figure 5

Sensitivity and Specificity of Human Expert Diagnosis of IPF

Full size image

UIP

ML model

In the diagnosis of UIP, the combined C-index, sensitivity, and specificity of ML were 0.91 (95% CI 0.87–0.94), 0.92 (95% CI 0.80–0.97) and 0.92 (95% CI 0.82–0.97), respectively. Notably, Aya Fukushima's deep learning model exhibits the highest diagnostic performance, boasting a C-index of 0.99 (95% CI 0.97 ~ 1.02), as illustrated in Figs. 6 and 7A.

Fig. 6
figure 6

Forest Map of Meta-Analysis of C-index for Machine Learning in Diagnosis of UIP

Full size image
Fig. 7
figure 7

Forest Map of Meta-Analysis of Sensitivity and Specificity for Machine Learning in Diagnosis of UIP. (A Meta-Analysis of the Sensitivity and Specificity of Machine Learning in Diagnosis of UIP. B Sensitivity and Specificity of Human Expert Diagnosis of UIP)

Full size image

We utilized funnel plot analysis to investigate publication bias concerning machine learning diagnoses of UIP, and the findings from the funnel plot along with Egger's test reveal a significant publication bias within each validation set of the models (P < 0.05), as illustrated in Fig. 8.

Fig. 8
figure 8

Funnel plot of Meta-Analysis of C-index for Machine Learning in Diagnosis of UIP

Full size image
Radiologists/clinicians

Ten pieces of literature did not explore the joint use of doctors and ML models in the diagnosis of UIP. The diagnostic sensitivity of radiologists/clinicians based solely on HRCT was reported to be 0.69 (95% CI 0.50 ~ 0.84), which was comparable to the diagnostic sensitivity of ML models, as depicted in Fig. 7B.

Discussion

Summary of the main findings

We have observed that the primary source of IPF for ML remains HRCT, with only a limited number of studies utilizing chest X-rays. In this study, specifically, only one study employed chest X-rays as a variable [21], making direct comparisons with CT challenging. The ML methods employed include convolutional neural networks, deep learning, and random forest. In the diagnosis of IPF, ML demonstrates superior sensitivity and specificity compared to clinicians/radiologists. For the diagnosis of UIP, whether examining the UIP pattern on HRCT or the pathological UIP pattern, variables are consistently derived from HRCT. The primary ML methods include convolutional neural networks, and deep learning, with a smaller representation of extreme gradient boosting (XGboost). In UIP diagnosis, the diagnostic sensitivity and specificity of ML align closely with those of clinicians/radiologists.

Comparison with previous studies

IPF needs to be distinguished not only from various interstitial lung disease (ILD) but also from specific types of IIP. Existing research indicates that IPF comprises 47–71% of IIP cases, whereas non-specific Interstitial Pneumonia (NSIP) accounts for 13–30% of IIP cases [34, 35]. The subjects in five included studies are patients with ILD or diffuse lung lesions. The primary outcome indicators focus on the sensitivity, specificity, and diagnostic performance of ML in the detection of IPF among various ILDs. Among these studies, three [20,21,22] compared ML with human experts, revealing significantly lower diagnostic sensitivity of human experts compared to ML (Fig. 7B). The study by Raghu [12] found that, despite all 95 cases being confirmed as IPF by surgical lung biopsy, interstitial disease experts' sensitivity in diagnosing IPF based on CT features was 78.5%. Another study [36, 37] reported low sensitivity of human experts in the diagnosis of interstitial pneumonia or interstitial pulmonary fibrosis combined with emphysema. Therefore, HRCT-based deep learning is deemed meaningful for diagnosing IPF. Its pooled diagnostic sensitivity is higher than that of clinical/radiological experts, potentially reducing interobserver variability and shortening the time from suspicion to diagnosis. It proves advantageous in the detection of IPF among various ILDs.

Regarding the diagnosis of UIP, the IPF Guidelines published in 2011 emphasized the role of identifying HRCT manifestations, including UIP patterns, as one of the independent diagnostic criteria. Studies comparing pathology with HRCT confirmed the accuracy of HRCT diagnosis of classic UIP at 80–90%. However, in some early UIP diagnosis studies, the diagnostic specificity was relatively low, at 43–78%. This decrease in specificity is related to patients with no honeycombing or atypical features on CT, limiting radiologists' ability to diagnose UIP based solely on CT. Among the studies included in this analysis, the pooled sensitivity of ML for diagnosing UIP is 92% (95% CI 0.80 ~ 0.97), with a specificity of 92% (95% CI 0.82 ~ 0.97). The sensitivity and diagnostic performance of ML in distinguishing UIP from non-UIP are comparable to those of human experts. In the future, ML algorithms may have an advantage in diagnosing potential UIP and non-UIP patterns.

Types of ML

Common ML models are applicable to interpretable clinical features, and some researchers have conducted ML studies based on imageological data. However, in the implementation process, manual image region segmentation and texture feature extraction are required, which may introduce bias due to the influence of the observer's prior knowledge. Deep learning is primarily applied to the identification of medical images, diagnosis, and prognosis of diseases, demonstrating a higher diagnostic rate. In our study, we included studies on both ML and deep learning. Due to the limited number of included studies, we did not conduct separate meta-analyses for ML and deep learning, but only briefly reviewed them. Therefore, more studies are desired to explore whether ML methods constructed based on conventional interpretable clinical features and images will outperform deep learning. Another advantage of deep learning is its capacity to intelligently diagnose diseases, providing crucial insights for the development of intelligent tools.

Advantages and limitations of the study

The strength of our study lies in being the first systematic review of the diagnostic value of ML in IPF and UIP, providing a comprehensive summary of diagnostic results compared with clinical/radiological findings. However, certain limitations should be acknowledged. Firstly, although a comprehensive systematic search was conducted, the number of studies included remains relatively low, which calls for careful interpretation of the findings. Secondly, the sensitivity analysis indicated that the predictive performance was not significantly influenced by chest radiography, which was one of the variables alongside others from high-resolution CT. Thirdly, during the image acquisition process, the image parameters were not modified, and preliminary experiments were carried out with varying image parameters. Consequently, it was not feasible to eliminate the heterogeneity resulting from the transitional setup of the equipment. Lastly, high heterogeneity presents a considerable challenge in machine learning-based meta-analysis. Given the restricted number of included studies, we were unable to further investigate this concern.

Conclusions

This study systematically evaluated the diagnostic performance of machine learning (ML) based on high-resolution CT imaging for idiopathic pulmonary fibrosis (IPF) and usual interstitial pneumonia (UIP). The results demonstrate that ML models leverage their strengths in data processing and pattern recognition to achieve rapid and efficient disease classification with high sensitivity and specificity. This highlights the potential of AI-based imaging diagnostic tools in enhancing early disease screening efficiency and reducing human error.

From a clinical standpoint, machine learning models can proficiently tackle the observer variability that is inherent in conventional diagnostic methods. Specifically, they can improve both consistency and accuracy in multidisciplinary collaborative diagnoses. Furthermore, this study conducts a systematic comparison between machine learning and traditional clinical radiological diagnostic approaches, validating the capability of machine learning to enhance diagnostic efficiency for IPF and UIP. This could serve as a foundation for its incorporation into future clinical practice. Given these findings, prospective clinical applications might include the integration of machine learning algorithms into hospital imaging analysis systems, facilitating expedited and more precise early screening for pulmonary fibrosis, minimizing diagnostic cycles, and ultimately enhancing patient outcomes.

Nevertheless, the utilization of these technologies necessitates additional extensive models and multicenter clinical trials to further confirm their applicability and effectiveness in diverse clinical environments. Consequently, although the existing findings underscore the considerable promise of machine learning, its broad implementation in clinical practice still demands thoughtful advancement.

Availability of data and materials

The data supporting the results in this study are available on request from the first author (Li Cong).

References

  1. Maher TM, Bendstrup E, Dron L, et al. Global incidence and prevalence of idiopathic pulmonary fibrosis. Respir Res. 2021;22:197. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12931-021-01791-z.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Riddell P, Kleinerova J, Eaton D, et al. Meaningful survival benefit for single lung transplantation in idiopathic pulmonary fibrosis patients over 65 years of age. Eur Respir J. 2020. https://doiorg.publicaciones.saludcastillayleon.es/10.1183/13993003.02413-2019.

    Article  PubMed  Google Scholar 

  3. Majewski S, Królikowska M, Costabel U, et al. Double lung transplantation for idiopathic pulmonary fibrosis in a patient with a history of liver transplantation and prolonged journey for disease-specific antifibrotic therapy. Case Rep Pulmonol. 2022;2022:4054339. https://doiorg.publicaciones.saludcastillayleon.es/10.1155/2022/4054339.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Hyldgaard C, Møller J, Bendstrup E. Changes in management of idiopathic pulmonary fibrosis: impact on disease severity and mortality. Eur Clin Respir J. 2020;7:1807682. https://doiorg.publicaciones.saludcastillayleon.es/10.1080/20018525.2020.1807682.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Spencer LG, Loughenbury M, Chaudhuri N, et al. Idiopathic pulmonary fibrosis in the UK: analysis of the British thoracic society electronic registry between 2013 and 2019. ERJ Open Res. 2021. https://doiorg.publicaciones.saludcastillayleon.es/10.1183/23120541.00187-2020.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Aono Y, Nakamura Y, Kono M, et al. Prognostic significance of forced vital capacity decline prior to and following antifibrotic therapy in idiopathic pulmonary fibrosis. Ther Adv Respir Dis. 2020. https://doiorg.publicaciones.saludcastillayleon.es/10.1177/1753466620953783.

    Article  PubMed  PubMed Central  Google Scholar 

  7. Raghu G, Collard HR, Egan JJ, et al. An official ATS/ERS/JRS/ALAT statement: idiopathic pulmonary fibrosis: evidence-based guidelines for diagnosis and management. Am J Respir Crit Care Med. 2011;183:788–824. https://doiorg.publicaciones.saludcastillayleon.es/10.1164/rccm.2009-040GL.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Widell J, Lidén M. Interobserver variability in high-resolution CT of the lungs. Eur J Radiol Open. 2020;7:100228. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.ejro.2020.100228.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Camp R, Smith ML, Larsen BT, et al. Reliability of histopathologic diagnosis of fibrotic interstitial lung disease: an international collaborative standardization project. BMC Pulm Med. 2021;21:184. https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s12890-021-01522-6.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Guiot J, Vaidyanathan A, Deprez L, et al. A review in radiomics: making personalized medicine a reality via routine imaging. Med Res Rev. 2022;42:426–40. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/med.21846.

    Article  PubMed  Google Scholar 

  11. Cho YH, Seo JB, Lee SM, et al. Quantitative CT imaging in chronic obstructive pulmonary disease: review of current status and future challenges. J Korean Soc Radiol. 2018;78:1–12. https://doiorg.publicaciones.saludcastillayleon.es/10.3348/jksr.2018.78.1.1.

    Article  Google Scholar 

  12. Tominaga J, Sakai F, Johkoh T, et al. Diagnostic certainty of idiopathic pulmonary fibrosis/usual interstitial pneumonia: the effect of the integrated clinico-radiological assessment. Eur J Radiol. 2015;84:2640–5. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.ejrad.2015.08.016.

    Article  PubMed  Google Scholar 

  13. Hochhegger B, Marchiori E, Zanon M, et al. Imaging in idiopathic pulmonary fibrosis: diagnosis and mimics. Clinics. 2019;74:e225. https://doiorg.publicaciones.saludcastillayleon.es/10.6061/clinics/2019/e225.

    Article  PubMed  PubMed Central  Google Scholar 

  14. Takahashi R, Kajikawa Y. Computer-aided diagnosis: a survey with bibliometric analysis. Int J Med Inform. 2017;101:58–67. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.ijmedinf.2017.02.004.

    Article  PubMed  Google Scholar 

  15. Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism. 2017;69:S36-s40. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.metabol.2017.01.011.

    Article  CAS  Google Scholar 

  16. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521:436–44. https://doiorg.publicaciones.saludcastillayleon.es/10.1038/nature14539.

    Article  CAS  PubMed  Google Scholar 

  17. Debray TP, Damen JA, Riley RD, et al. A framework for meta-analysis of prediction model studies with binary and time-to-event outcomes. Stat Methods Med Res. 2019. https://doiorg.publicaciones.saludcastillayleon.es/10.1177/0962280218785504.

    Article  PubMed  Google Scholar 

  18. Yu W, Zhou H, Goldin JG, et al. End-to-end domain knowledge-assisted automatic diagnosis of idiopathic pulmonary fibrosis (IPF) using computed tomography (CT). Med Phys. 2021;48:2458–67. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/mp.14754.

    Article  PubMed  Google Scholar 

  19. Yu W, Zhou H, Choi Y, et al. Multi-scale, domain knowledge-guided attention + random forest: a two-stage deep learning-based multi-scale guided attention models to diagnose idiopathic pulmonary fibrosis from computed tomography images. Med Phys. 2023;50:894–905. https://doiorg.publicaciones.saludcastillayleon.es/10.1002/mp.16053.

    Article  PubMed  Google Scholar 

  20. Refaee T, Salahuddin Z, Frix AN, et al. Diagnosis of idiopathic pulmonary fibrosis in high-resolution computed tomography scans using a combination of handcrafted radiomics and deep learning. Front Med. 2022;9:915243. https://doiorg.publicaciones.saludcastillayleon.es/10.3389/fmed.2022.915243.

    Article  Google Scholar 

  21. Abe H, Ashizawa K, Li F, et al. Artificial neural networks (ANNs) for differential diagnosis of interstitial lung disease: results of a simulation test with actual clinical cases. Acad Radiol. 2004;11:29–37. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/s1076-6332(03)00572-5.

    Article  PubMed  Google Scholar 

  22. Fukushima A, Ashizawa K, Yamaguchi T, et al. Application of an artificial neural network to high-resolution CT: usefulness in differential diagnosis of diffuse lung disease. AJR Am J Roentgenol. 2004;183:297–305. https://doiorg.publicaciones.saludcastillayleon.es/10.2214/ajr.183.2.1830297.

    Article  PubMed  Google Scholar 

  23. Walsh SLF, Calandriello L, Silva M, et al. Deep learning for classifying fibrotic lung disease on high-resolution computed tomography: a case-cohort study. Lancet Respir Med. 2018;6:837–45. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/s2213-2600(18)30286-8.

    Article  PubMed  Google Scholar 

  24. Bratt A, Williams JM, Liu G, et al. Predicting usual interstitial pneumonia histopathology from chest CT imaging with deep learning. Chest. 2022;162:815–23. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.chest.2022.03.044.

    Article  PubMed  Google Scholar 

  25. Christe A, Peters AA, Drakopoulos D, et al. Computer-aided diagnosis of pulmonary fibrosis using deep learning and CT images. Invest Radiol. 2019;54:627–32. https://doiorg.publicaciones.saludcastillayleon.es/10.1097/rli.0000000000000574.

    Article  PubMed  PubMed Central  Google Scholar 

  26. Depeursinge A, Chin AS, Leung AN, et al. Automated classification of usual interstitial pneumonia using regional volumetric texture analysis in high-resolution computed tomography. Invest Radiol. 2015;50:261–7. https://doiorg.publicaciones.saludcastillayleon.es/10.1097/rli.0000000000000127.

    Article  PubMed  PubMed Central  Google Scholar 

  27. Koo CW, Williams JM, Liu G, et al. Quantitative CT and machine learning classification of fibrotic interstitial lung diseases. Eur Radiol. 2022;32:8152–61. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s00330-022-08875-4.

    Article  PubMed  Google Scholar 

  28. Shaish H, Ahmed FS, Lederer D, et al. Deep learning of computed tomography virtual wedge resection for prediction of histologic usual interstitial pneumonitis. Ann Am Thorac Soc. 2021;18:51–9. https://doiorg.publicaciones.saludcastillayleon.es/10.1513/AnnalsATS.202001-068OC.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Humphries SM, Thieke D, Baraghoshi D, et al. Deep learning classification of usual interstitial pneumonia predicts outcomes. Am J Respir Crit Care Med. 2024;209:1121–31. https://doiorg.publicaciones.saludcastillayleon.es/10.1164/rccm.202307-1191OC.

    Article  PubMed  Google Scholar 

  30. Chang M, Reicher JJ, Kalra A, et al. Analysis of validation performance of a machine learning classifier in interstitial lung disease cases without definite or probable usual interstitial pneumonia pattern on CT using clinical and pathology-supported diagnostic labels. J Imaging Inform Med. 2024;37:297–307. https://doiorg.publicaciones.saludcastillayleon.es/10.1007/s10278-023-00914-w.

    Article  PubMed  PubMed Central  Google Scholar 

  31. Fan W, Chen Q, Maccarrone V, et al. Developing radiology diagnostic tools for pulmonary fibrosis using machine learning methods. Clin Imaging. 2024;106:110047. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.clinimag.2023.110047.

    Article  PubMed  Google Scholar 

  32. Chung JH, Chelala L, Pugashetti JV, et al. A deep learning-based radiomic classifier for usual interstitial pneumonia. Chest. 2024;165:371–80. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.chest.2023.10.012.

    Article  PubMed  Google Scholar 

  33. Maddali MV, Kalra A, Muelly M, et al. Development and validation of a CT-based deep learning algorithm to augment non-invasive diagnosis of idiopathic pulmonary fibrosis. Respir Med. 2023;219:107428. https://doiorg.publicaciones.saludcastillayleon.es/10.1016/j.rmed.2023.107428.

    Article  PubMed  Google Scholar 

  34. Travis WD, Hunninghake G, King TE Jr, et al. Idiopathic nonspecific interstitial pneumonia: report of an American thoracic society project. Am J Respir Crit Care Med. 2008;177:1338–47. https://doiorg.publicaciones.saludcastillayleon.es/10.1164/rccm.200611-1685OC.

    Article  PubMed  Google Scholar 

  35. Monaghan H, Wells AU, Colby TV, et al. Prognostic implications of histologic patterns in multiple surgical lung biopsies from patients with idiopathic interstitial pneumonias. Chest. 2004;125:522–6. https://doiorg.publicaciones.saludcastillayleon.es/10.1378/chest.125.2.522.

    Article  PubMed  Google Scholar 

  36. Reddy TL, Tominaga M, Hansell DM, et al. Pleuroparenchymal fibroelastosis: a spectrum of histopathological and imaging phenotypes. Eur Respir J. 2012;40:377–85. https://doiorg.publicaciones.saludcastillayleon.es/10.1183/09031936.00165111.

    Article  PubMed  Google Scholar 

  37. Kheir F, Uribe Becerra JP, Bissell B, et al. Use of a genomic classifier in patients with interstitial lung disease: a systematic review and meta-analysis. Ann Am Thorac Soc. 2022;19:827–32. https://doiorg.publicaciones.saludcastillayleon.es/10.1513/AnnalsATS.202102-197OC.

    Article  PubMed  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

This work was supported by the Immunological mechanism of stem cell therapy for fibrotic hypersensitivity pneumonia (2022TSYCCX0036). The funding had no role in the study’s design, conduct, or reporting and could neither approve nor disapprove the submitted manuscript.

Author information

Author notes

  1. Li Cong and Ying Chen have contributed equally to this work.

Authors and Affiliations

  1. Respiratory and Critical Care Medical Center, People’s Hospital of Xinjiang Uygur Autonomous Region, Urumqi, 830000, China

    Li Cong, Ying Chen & Liping Chen

  2. Radiographic Center, People’s Hospital of Xinjiang Uygur Autonomous Region, Urumqi, China

    Xiaolei He & MaiLiKai KuErBan

  3. Department of Respiratory and Critical Care Medicine, The First Affiliated Hospital of Shihezi University, Shihezi, China

    Chao Wu

Authors
  1. Li Cong
    View author publications

    You can also search for this author inPubMed Google Scholar

  2. Ying Chen
    View author publications

    You can also search for this author inPubMed Google Scholar

  3. Xiaolei He
    View author publications

    You can also search for this author inPubMed Google Scholar

  4. MaiLiKai KuErBan
    View author publications

    You can also search for this author inPubMed Google Scholar

  5. Chao Wu
    View author publications

    You can also search for this author inPubMed Google Scholar

  6. Liping Chen
    View author publications

    You can also search for this author inPubMed Google Scholar

Contributions

LC: Data curation; Writing—original draft. YC: Data curation; Writing—original draft. XH: Formal analysis; Validation. Malika: Data curation; Formal analysis. CW: Data curation; Validation. LC: Conceptualization; Project administration; Supervision.

Corresponding author

Correspondence to Liping Chen.

Ethics declarations

Ethics approval and consent to participate

Institutional Review Board approval was not required because this study is based exclusively on published literature.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Cong, L., Chen, Y., He, X. et al. Diagnosis accuracy of machine learning for idiopathic pulmonary fibrosis: a systematic review and meta-analysis. Eur J Med Res 30, 288 (2025). https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s40001-025-02501-x

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doiorg.publicaciones.saludcastillayleon.es/10.1186/s40001-025-02501-x

Keywords

European Journal of Medical Research

ISSN: 2047-783X

Contact us