Application of Artificial Intelligence in Idiopathic Pulmonary Fibrosis: History
View Latest Version
Please note this is an old version of this entry, which may differ significantly from the current revision.
Contributor:
Gerui Zhang
,
Lin Luo
,
Limin Zhang
,
Zhuo Liu

Machine Learning (ML) is an algorithm based on big data, which learns patterns from the previously observed data through classifying, predicting, and optimizing to accomplish specific tasks.  ML can have good performance and is a great potential tool, especially in the imaging diagnosis of interstitial lung disease.  Idiopathic pulmonary fibrosis (IPF) is a major problem in the treatment of respiratory diseases, due to the abnormal proliferation of fibroblasts, leading to lung tissue destruction. The diagnosis mainly depends on the early detection of imaging and early treatment, which can effectively prolong the life of patients. 

  • interstitial pulmonary fibrosis
  • machine learning
  • artificial intelligence

1. Introduction

Machine learning (ML) is a subfield of artificial Intelligence (AI) and is based on a big-data-based algorithm that classifies, predicts, and optimizes according to previously observed data, using data to identify trends and complete specified tasks. ML contains two types of learning: supervised learning and unsupervised learning, and the range of techniques has gradually developed from simple linear models for complex neural networks with a large number of parameters. Multiple layers of “neurons” make up artificial neural networks (ANNs), which are based on the human brain and continuously process input data until they reach the output layer. Deep learning (DL), also known as convolutional neural networks (CNNs), is a recently developed variant of ANN that outputs data in a hierarchical manner, with successive layers evolving in between, processing incoming data in a fashion that includes both abstract high-level qualities, such as distinct objects, and simple low-level features, such as linearity [1].
A paradigm change in artificial intelligence is present along with CNN. In the early stages of AI research, the aim was to incorporate supervised learning into rule-based “expert systems” that could classify chest radiograph images as “normal” or “abnormal”. CNNs can more quickly distinguish between data availability and accuracy from large training datasets, for example, as the extensive usage of picture archiving and communication systems and electronic health records (EHRs). As an auxiliary tool for clinicians, machine learning has developed rapidly in medicine, especially in the application of respiratory diseases. Pulmonary imaging analysis can help distinguish normal lung tissue from ground glass opacities and honeycomb-like lung tissue changes, and assist in the differentiation of benign and malignant pulmonary nodules. Machine learning can assist in assessing indications for mechanical ventilation and the timing of weaning. In chronic respiratory diseases, it can assist in the assessment of pulmonary function to predict prognosis and treatment effect. In terms of respiratory biological information monitoring, it can help to monitor the early diagnosis of obstructive sleep apnea syndrome and reduce the occurrence of complications.
Early lung imaging of idiopathic pulmonary fibrosis (IPF) lacks evident specificity; thus, the accuracy of the diagnosis depends on the appropriate high-precision radiological imaging technology and is also constrained by the experience and expertise of radiologists and doctors. IPF is a chronic progressive inflammatory disease caused by a variety of reasons, with diffuse pulmonary parenchyma, alveolar inflammation, and interstitial fibrosis as the basic pathological lesions. In particular, the main clinical diagnostic methods of IPF include pulmonary imaging examination, lung biopsy, and pulmonary function test [2][3][4]. If the computer can help with the findings of an examination for fibrosis, it will aid in the early detection of such diseases, which is very beneficial to both patients and medical professionals.
Clinically, the situation of chest X-ray alone is complex, and it is difficult to effectively diagnose fibrotic lesions. High-resolution CT (HRCT) is usually characterized by “nonspecific interstitial pneumonia”, which shows honeycomb-like or stretch bronchiectasis or bronchiectasis in the bilateral subpleural base. Peripheral opaque ground-glass changes were most prominent in small nodules in the lower lobes. However, these are atypical lesions, which need to be differentiated from a variety of clinical disease-related factors, to further exclude autoimmune or drug factors, and exclude diseases other than interstitial pneumonia.

2. Application of Artificial Intelligence in the Respiratory System

2.1. Imaging Analysis of Pulmonary Nodules

Imaging analysis plays an integral role in the diagnosis and treatment of pulmonary diseases. DL and CNN are mainly used in medical imaging and have achieved promising results in lung nodule detection, as well as excellent performance in segmentation and classification of pulmonary nodules [5]. According to Siegel et al. [6], the 5-year survival rate is exceedingly dismal, and 55% of lung cancer patients have distant metastases at the time of initial diagnosis. Therefore, accurate classification and diagnosis of pulmonary nodules are essential to reduce the morbidity and mortality of early lung cancer. During image processing, CNN segments the image and isolates the analyzed object from the surrounding environment for analysis to evaluate the nodule size as a predictor of benign or malignant tumor. The volume method evaluates the sensitivity of nodule growth by reproducing and 3D-analyzing the size detection of nodules, and it is now regarded as the best technique for determining nodule size and growth [7].

2.2. Application of Racial Intelligence in Respiratory Monitoring in Critical Care Medicine

Mechanical ventilation is an important area of intensive care units (ICUs). It is a lifesaving tool that provides respiratory support to patients with respiratory failure in the ICU, and it is the focus of research in ML. Inappropriate mechanical ventilation may worsen lung injury, prolong the duration of mechanical ventilation, increase the risk of infection, and increase mortality. By collecting clinical parameters and laboratory results of critically ill patients, ML helps clinicians to predict the necessity of intubation within 24 h of admission to critically ill patients [8].
Prediction of infectious etiology is also an important direction of ML research. Sepsis and septic shock in severe infection are also one of the major life-threatening problems in intensive care. In many stages of sepsis, including early detection, prognostic assessment, mortality prediction, and clinical management, machine learning methods can be applied. Every hour of delay increases patient mortality, making early care in sepsis crucial. Initial sepsis prediction systems relied heavily on empirical clinical decision rules (CDRS), often using vital signs collected at the bedside. 

2.3. Artificial Intelligence in Lung Function Assessment and Prediction of Chronic Respiratory Diseases

Chronic airway diseases mainly refer to asthma and chronic obstructive pulmonary disease, and the incidence and economic burden of developing countries are among the highest in the world [9]. It is characterized by ongoing inflammation, airway remodeling, obstruction, and recurrence, which significantly lowers quality of life and raises the possibility of hospitalization and death. For accurate prevention and individualized therapy, asthma is a diverse illness with numerous phenotypes and genotypes that must be appropriately differentiated. In recent years, various ML algorithms have employed genetic data in conjunction with clinical information, such as laboratory test results, to identify asthma phenotypes [10]. Spirometry and bronchial provocation tests, as well as eosinophil count analysis and fractional exhaled nitric oxide measurement, are employed in clinical practice to evaluate airflow restriction and hyperresponsiveness, allowing the identification of certain asthma phenotypes [11].

3. Application of Artificial Intelligence in IPF

IPF is a chronic progressive destruction of the lung disease. The average survival time is less than 5 years, and the early clinical manifestations of patients are not obvious, mainly manifesting as shortness of breath after activity, dry cough, recurrent lower respiratory tract infection, no obvious specificity, in the middle and late stage of progressive dyspnea, irreversible respiratory failure, and eventually death. The pulmonary function of the patients is delayed, usually with restrictive ventilation dysfunction, especially the reduction of forced vital capacity, total lung capacity, functional residual capacity, and diffusion capacity of the lungs for carbon monoxide (DLCO) [12]. Imaging plays a crucial role in the diagnosis of IFP. Chest X-ray is usually used as the initial means of imaging diagnosis of interstitial lung disease, but in some underdeveloped areas, chest X-ray is an indispensable part of imaging evaluation when critical patients are examined at the bedside. As a more sensitive imaging technique, high-resolution computed tomography (HRCT) is considered to be the core diagnostic tool for interstitial lung disease [13]. Abnormal lesions, such as irregular linear shadows, honeycombing, and reticular changes, observed on continuous HRCT can help radiologists and clinicians to identify specific interstitial lung disease lesions and assess the progression of the disease. However, at present, the assessment of the progression of interstitial lung disease mostly relies on the doctor’s visual analysis, which has certain subjective factors. Due to the limitation of clinical level, its accuracy and sensitivity are low. The detection rate of bronchoscopic lung biopsy for interstitial lung disease is also extremely low, and surgical lung biopsy is undoubtedly the gold standard for diagnosis. However, surgical lung biopsy has certain risks. For young patients with good lung function tolerance, it may have a certain guiding effect on their future treatment and prognosis. The risks of lung biopsy are so great that it may even increase the risk of death.
Deep learning approaches are used to identify, categorize, and segment ILD pictures on HRCT. At present, the most commonly used interstitial fibrosis mode CNN segmentation is U-Net [14]. Park et al. [15] analyzed 647 patients with lung segmentation by HRCT in ILD, and the accuracy reached 98%. Data augmentation is used in image processing to increase the amount of training data available. Common operations include image flipping, rotation, cropping, and scaling. Combined with data enhancement, the accuracy of fibrosis morphological classification can be improved to 78–91%. Using the 2011 ATS/ERS/JRS/ALAT criteria and the 2018 Fleischner Society criteria [12][16], Walsh et al. [17] developed a model to classify 1307 HRCT images of pulmonary fibrosis. Use of the UIP-HRCT model for pathological classification of fibrosis tissue can avoid the need for lung biopsy to a certain extent. Two studies [17][18] showed good performance in the diagnosis of IPF, which was close to expert level, and the diagnostic accuracy of the research tool was 78.9%, but there is also a high risk of bias due to the limited number of retrospective studies. Therefore, the value of current artificial intelligence technologies for the evaluation of ILD can only be reliably assessed by well-designed prospective controlled trials, and better evaluation algorithms and tools need to be further developed. Sikandar et al. [19] developed and trained the Forest model to evaluate 2424 subjects to predict the severity of pulmonary fibrosis patients, and the model achieved sensitivity and accuracy of 0.71 and 0.64, respectively. This model will help clinicians to diagnose IPF patients and assess the severity of the disease at an early stage, and make timely positive measures related to the treatment of IPF.
It has been demonstrated that certain CT-sensitive characteristics, such as reticulation and honeycomb, can accurately predict death in IPF patients. However, as prolonged progressive fibrotic ILD death is frequently unachievable End points, such as IPF, have the potential to be used as a substitute for therapy evaluation by changing the extent of the disease on high-resolution CT. In response, CALIPER has proven useful for tracking and forecasting disease [20][21][22]. Emphysema estimations based on threshold algorithms in a recent CALIPER-derived CT study [23] were considerably impacted by radiation dose, whereas the impact of dose reduction on texture-based algorithms was less thoroughly studied. There was a substantial association between CALIPER-derived CT characteristics and lung function results (FVC and FEV1%) in patients receiving treatment for pulmonary fibrosis, which was the first to assess the impact of CT dosage adjustments on CALIPER performance. These results are in line with those of earlier retrospective investigations, but bigger prospective studies with longer follow-up times would undoubtedly be required to confirm the present results. Similar to PFT, CALIPER may be impacted by variations in lung volume; hence, it is crucial to have an experienced radiologist double-check the data. The use of conventional machine learning quantification software may be seen as a limitation given the increasing availability and complexity of deep learning algorithms; however, the datasets necessary to train such algorithms would require hundreds of thousands of cases, a difficult threshold to reach given the relatively low prevalence of this disease in the general population. When evaluating pulmonary fibrosis in IPF patients taking antifibrosis therapy, CALIPER parameters corresponded well with lung function, and CT dosage decrease had no effect on the software’s performance. When evaluating pulmonary fibrosis in treated patients, CALIPER can be a potent and objective addition to traditional lung function proxies in the absence of confounding factors affecting lung function.

This entry is adapted from the peer-reviewed paper 10.3390/diagnostics13030357

References

  1. LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444.
  2. Yang, Y.; Ni, J.; Xu, B.; Luo, G. Progress in study of comorbidities in idiopathic pulmonary fibrosis. Int. J. Respir. 2021, 41, 690–695.
  3. Baroke, E.; Gauldie, J.; Kolb, M. New treatment and markers of prognosis for idiopathic pulmonary fibrosis: Lessons learned from translational research. Expert Rev. Respir. Med. 2013, 7, 465–478.
  4. Rochwerg, B.; Neupane, B.; Zhang, Y.; Garcia, C.C.; Raghu, G.; Richeldi, L.; Brozek, J.; Beyene, J.; Schünemann, H. Treatment of idiopathic pulmonary fibrosis: A network meta-analysis. BMC Med. 2016, 14, 18.
  5. Tandon, Y.K.; Bartholmai, B.J.; Koo, C.W. Putting artificial intelligence (AI) on the spot: Machine learning evaluation of pulmonary nodules. J. Thorac. Dis. 2020, 12, 6954–6965.
  6. Siegel, R.L.; Miller, K.D.; Jemal, A. Cancer statistics, 2019. CA Cancer J. Clin. 2019, 69, 7–34.
  7. Han, D.; Heuvelmans, M.A.; Vliegenthart, R.; Rook, M.; Dorrius, M.D.; De Jonge, G.J.; Walter, J.E.; Van Ooijen, P.M.A.; De Koning, H.J.; Oudkerk, M. Influence of lung nodule margin on volume- and diameter-based reader variability in CT lung cancer screening. Br. J. Radiol. 2018, 91, 20170405.
  8. Siu, B.M.K.; Kwak, G.H.; Ling, L.; Hui, P. Predicting the need for intubation in the first 24 h after critical care admission using machine learning approaches. Sci. Rep. 2020, 10, 20931.
  9. Feng, Y.; Wang, Y.; Zeng, C.; Mao, H. Artificial Intelligence and Machine Learning in Chronic Airway Diseases: Focus on Asthma and Chronic Obstructive Pulmonary Disease. Int. J. Med. Sci. 2021, 18, 2871–2889.
  10. Krautenbacher, N.; Flach, N.; Böck, A.; Laubhahn, K.; Laimighofer, M.; Theis, F.J.; Ankerst, D.P.; Fuchs, C.; Schaub, B. A strategy for high-dimensional multivariable analysis classifies childhood asthma phenotypes from genetic, immunological, and environmental factors. Allergy 2019, 74, 1364–1373.
  11. Korevaar, D.A.; Westerhof, G.A.; Wang, J.; Cohen, J.F.; Spijker, R.; Sterk, P.J.; Bel, E.H.; Bossuyt, P.M.M. Diagnostic accuracy of minimally invasive markers for detection of airway eosinophilia in asthma: A systematic review and meta-analysis. Lancet Respir. Med. 2015, 3, 290–300.
  12. Lynch, D.A.; Sverzellati, N.; Travis, W.D.; Brown, K.K.; Colby, T.V.; Galvin, J.R.; Goldin, J.G.; Hansell, D.M.; Inoue, Y.; Johkoh, T.; et al. Diagnostic criteria for idiopathic pulmonary fibrosis: A Fleischner Society White Paper. Lancet Respir. Med. 2018, 6, 138–153.
  13. Soffer, S.; Morgenthau, A.S.; Shimon, O.; Barash, Y.; Konen, E.; Glicksberg, B.S.; Klang, E. Artificial Intelligence for Interstitial Lung Disease Analysis on Chest Computed Tomography: A Systematic Review. Acad. Radiol. 2021, 29, S226–S235.
  14. Ronneberger, O.; Fischer, P.; Brox, T. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention; Springer: Berlin/Heidelberg, Germany, 2015; pp. 234–241.
  15. Park, B.; Park, H.; Lee, S.M.; Seo, J.B.; Kim, N. Lung Segmentation on HRCT and Volumetric CT for Diffuse Interstitial Lung Disease Using Deep Convolutional Neural Networks. J. Digit. Imaging 2019, 32, 1019–1026.
  16. Raghu, G.; Remy-Jardin, M.; Myers, J.L.; Richeldi, L.; Ryerson, C.J.; Lederer, D.J.; Behr, J.; Cottin, V.; Danoff, S.K.; Morell, F.; et al. Diagnosis of Idiopathic Pulmonary Fibrosis. An Official ATS/ERS/JRS/ALAT Clinical Practice Guideline. Am. J. Respir. Crit. Care Med. 2018, 198, e44–e68.
  17. Walsh, S.L.F.; Calandriello, L.; Silva, M.; Sverzellati, N. Deep learning for classifying fibrotic lung disease on high-resolution computed tomography: A case-cohort study. Lancet Respir. Med. 2018, 6, 837–845.
  18. Christe, A.; Peters, A.A.; Drakopoulos, D.; Heverhagen, J.; Geiser, T.; Stathopoulou, T.; Christodoulidis, S.; Anthimopoulos, M.; Mougiakakou, S.G.; Ebner, L. Computer-Aided Diagnosis of Pulmonary Fibrosis Using Deep Learning and CT Images. Investig. Radiol. 2019, 54, 627–632.
  19. Ali, S.; Hussain, A.; Aich, S.; Park, M.S.; Chung, M.P.; Jeong, S.H.; Song, J.W.; Lee, J.H.; Kim, H.C. A Soft Voting Ensemble-Based Model for the Early Prediction of Idiopathic Pulmonary Fibrosis (IPF) Disease Severity in Lungs Disease Patients. Life 2021, 11, 1092.
  20. Romei, C.; Tavanti, L.M.; Taliani, A.; De Liperi, A.; Karwoski, R.; Celi, A.; Palla, A.; Bartholmai, B.J.; Falaschi, F. Automated Computed Tomography analysis in the assessment of Idiopathic Pulmonary Fibrosis severity and progression. Eur. J. Radiol. 2020, 124, 108852.
  21. Maldonado, F.; Moua, T.; Rajagopalan, S.; Karwoski, R.A.; Raghunath, S.; Decker, P.A.; Hartman, T.E.; Bartholmai, B.; Robb, R.A.; Ryu, J. Automated quantification of radiological patterns predicts survival in idiopathic pulmonary fibrosis. Eur. Respir. J. 2013, 43, 204–212.
  22. Wu, X.; Kim, G.H.; Salisbury, M.L.; Barber, D.; Bartholmai, B.J.; Brown, K.K.; Conoscenti, C.S.; De Backer, J.; Flaherty, K.R.; Gruden, J.F.; et al. Computed Tomographic Biomarkers in Idiopathic Pulmonary Fibrosis: The Future of Quantitative Analysis. Am. J. Respir. Crit. Care Med. 2019, 199, 12–21.
  23. Koo, C.; Larson, N.; Parris-Skeete, C.; Karwoski, R.; Kalra, S.; Bartholmai, B.; Carmona, E. Prospective machine learning CT quantitative evaluation of idiopathic pulmonary fibrosis in patients undergoing anti-fibrotic treatment using low- and ultra-low-dose CT. Clin. Radiol. 2021, 77, e208–e214.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
This entry is offline, you can click here to edit this entry!
ScholarVision Creations
Feedback