Deep Learning for the Automatic Quantification of Pleural Plaques in Asbestos-Exposed Subjects

Our plain-English summary of Deep Learning for the Automatic Quantification of Pleural Plaques in Asbestos-Exposed Subjects, by I. Benlala, B.D. De Senneville, G. Dournes, et al (International Journal of Environmental Research and Public Health, 2022).

The study

• Authors: I. Benlala, B.D. De Senneville, G. Dournes, M. Menant, C. Gramond, I. Thaon, B. Clin, P. Brochard, A. Gislard, P. Andujar, S. Chammings, J. Gallet, A. Lacourt, F. Delva, C. Paris, G. Ferretti, J.C. Pairon, F. Laurent
• Journal: International Journal of Environmental Research and Public Health
• Published: 2022;19(3)
• Cited by: 2 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Objective: This study aimed to develop and validate an automated artificial intelligence (AI)-driven quantification of pleural plaques in a population of retired workers previously occupationally exposed to asbestos.

Methods: CT scans of former workers previously occupationally exposed to asbestos who participated in the multicenter APEXS (Asbestos PostExposure Survey) study were collected retrospectively between 2010 and 2017 during the second and the third rounds of the survey. A hundred and forty-one participants with pleural plaques identified by expert radiologists at the 2nd and the 3rd CT screenings were included. Maximum Intensity Projection (MIP) with 5 mm thickness was used to reduce the number of CT slices for manual delineation. A Deep Learning AI algorithm using 2D-convolutional neural networks was trained with 8280 images from 138 CT scans of 69 participants for the semantic labeling of Pleural Plaques (PP). In all, 2160 CT images from 36 CT scans of 18 participants were used for AI testing versus ground-truth labels (GT). The clinical validity of the method was evaluated longitudinally in 54 participants with pleural plaques.

Results: The concordance correlation coefficient (CCC) between AI-driven and GT was almost perfect (>0.98) for the volume extent of both PP and calcified PP. The 2D pixel similarity overlap of AI versus GT was good (DICE = 0.63) for PP, whether they were calcified or not, and very good (DICE = 0.82) for calcified PP. A longitudinal comparison of the volumetric extent of PP showed a significant increase in PP volumes (p < 0.001) between the 2nd and the 3rd CT screenings with an average delay of 5 years.

Conclusions: AI allows a fully automated volumetric quantification of pleural plaques showing volumetric progression of PP over a five-year period. The reproducible PP volume evaluation may enable further investigations for the comprehension of the unclear relationships between pleural plaques and both respiratory function and occurrence of thoracic malignancy.

In plain English

This French study tested a deep learning program that automatically measures pleural plaque volume on CT scans. It used scans of 141 retired workers, almost all men, exposed to asbestos at work, with pleural plaques at two French screening rounds (2010 and 2017). The program was trained on 69 participants and tested on 18 others against radiologists' hand-drawn outlines. In testing, volume concordance was 0.98 (95% CI 0.96 to 0.99) for all plaques and 0.99 (0.99 to 0.99) for calcified plaques; pixel overlap was 0.63 and 0.82. In a separate group of 54 participants, median plaque volume rose from 7.1 mL (95% CI 4.4 to 11.5) to 12.1 mL (9.8 to 16.9) over about five years, and calcified plaque volume from 1.3 mL (0.6 to 2.6) to 3.5 mL (2.2 to 5.3), both p less than 0.001. Median program time per scan was 18 seconds on processed (MIP) images and 90 seconds on native images, versus 15 minutes by hand. The authors concluded automated measurement is feasible and may help study how plaques relate to breathing function and chest cancer.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Retrospective cohort study developing and validating a deep learning (2D convolutional neural network) method for measuring pleural plaque volume on CT
• Who was studied: 141 retired workers in the French ARDCO surveillance program (APEXS CT study) with pleural plaques at both the second (2010) and third (2017) CT screening rounds. They were split into a training group (69 participants), a test group (18) and a clinical validation group (54). The groups were almost all male (68/1, 18/0 and 54/0 male/female).
• Exposure: Occupational asbestos exposure during working life. Fiber types are not specified. Median total exposure duration was 36, 38 and 35 years in the training, test and clinical validation groups.
• Main results:
  • Test group (2160 CT slices from 36 scans of 18 participants): pixel overlap (DICE) between the program and the hand-drawn outlines was 0.63 for pleural plaques and 0.82 for calcified pleural plaques.
  • Test group (36 scans): agreement on plaque volume (concordance correlation coefficient) was 0.98 (95% CI 0.96 to 0.99) for pleural plaques and 0.99 (0.99 to 0.99) for calcified plaques. The mean volume difference was less than 2.3 mL.
  • Program-measured plaque volume correlated with radiologists' visual extent score: rho = 0.66 in the test group (hand-drawn outlines rho = 0.68) and rho = 0.56 in the clinical validation group, all p less than 0.001.
  • Clinical validation group (54 participants): median pleural plaque volume rose from 7.1 mL (95% CI 4.4 to 11.5) to 12.1 mL (9.8 to 16.9), and calcified plaque volume from 1.3 mL (0.6 to 2.6) to 3.5 mL (2.2 to 5.3), between the 2nd and 3rd scans (p less than 0.001). The median increases were 5.71 mL (95% CI 3.57 to 9.48) and 2.85 mL (95% CI 1.62 to 7.76).
  • Clinical validation group: the median percentage increase was 70% for pleural plaques and 143% for calcified plaques.
  • Repeating the program gave DICE above 0.99. For hand outlining of 20 scans, DICE was 0.72 (all plaques) and 0.75 (calcified) between two observers, and 0.87 and 0.89 when the same observer repeated the work.
  • Manual outlining took a median of 15 minutes. The program took a median of 18 seconds on thin MIP images and 90 seconds on native images.
• The authors' conclusion: The authors concluded that automated AI measurement of pleural plaques is feasible. They said it may help researchers study the unclear relationships between pleural plaques and both breathing function and chest cancer.
• Limitations the authors state:
  • This was a retrospective first attempt to evaluate the method; the authors said large prospective studies may add more.
  • Detecting whether plaques are present was outside the study's scope; the method measures plaques already diagnosed by radiologists.
  • Scans came from four manufacturers; testing on other manufacturers and CT models would be needed to show the model works more generally.

Topics

Medical terms in this study, explained in our Lexicon: Pleural Plaques, CT Scan.

Related studies

• Interstitial Lung Abnormalities Detected by CT in Asbestos-Exposed Subjects Are More Likely Associated to Age — F. Laurent, I. Benlala, et al. (2021)
• Deep-learning based classification distinguishes sarcomatoid malignant mesotheliomas from benign spindle cell mesothelial proliferations — J.R. Naso, A.B. Levine, et al. (2021)
• Asbestos Detection with Fluorescence Microscopy Images and Deep Learning — C. Cai, T. Nishimura, et al. (2021)
• 3-Dimensional Quantification of Composite Pleural Plaque Volume in Patients Exposed to Asbestos Using High-resolution Computed Tomography: A Validation Study — G. Dournes, A. Dubois, et al. (2019)
• Relationships between asbestos exposure and pleural plaques: dose and time effects using fractional polynomials — M. Menant, I. Benlala, et al. (2024)
• Asbestos exposure, pleural plaques and digestive cancers — B. Clin, C. Gramond, et al. (2025)

Browse more research in International Journal of Environmental Research and Public Health, or return to medical research.