Deep learning with deep convolutional neural network using FDG-PET/CT for malignant pleural mesothelioma diagnosis

Our plain-English summary of Deep learning with deep convolutional neural network using FDG-PET/CT for malignant pleural mesothelioma diagnosis, by K. Kitajima, H. Matsuo, A. Kono, et al (Oncotarget, 2021).

The study

• Authors: K. Kitajima, H. Matsuo, A. Kono, K. Kuribayashi, T. Kijima, M. Hashimoto, S. Hasegawa, T. Murakami, K. Yamakado
• Journal: Oncotarget
• Published: 2021;12(12):1187-1196
• Cited by: 16 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Objectives: This study analyzed an artificial intelligence (AI) deep learning method with a three-dimensional deep convolutional neural network (3D DCNN) in regard to diagnostic accuracy to differentiate malignant pleural mesothelioma (MPM) from benign pleural disease using FDG-PET/CT results.

Results: For protocol A, the area under the ROC curve (AUC)/sensitivity/specificity/accuracy values were 0.825/77.9% (81/104)/76.4% (55/72)/77.3% (136/176), while those for protocol B were 0.854/80.8% (84/104)/77.8% (56/72)/79.5% (140/176), for protocol C were 0.881/85.6% (89/104)/75.0% (54/72)/81.3% (143/176), and for protocol D were 0.896/88.5% (92/104)/73.6% (53/72)/82.4% (145/176). Protocol D showed significantly better diagnostic performance as compared to A, B, and C in ROC analysis (p = 0.031, p = 0.0020, p = 0.041, respectively).

Materials And Methods: Eight hundred seventy-five consecutive patients with histologically proven or suspected MPM, shown by history, physical examination findings, and chest CT results, who underwent FDG-PET/CT examinations between 2007 and 2017 were investigated in a retrospective manner. There were 525 patients (314 MPM, 211 benign pleural disease) in the deep learning training set, 174 (102 MPM, 72 benign pleural disease) in the validation set, and 176 (104 MPM, 72 benign pleural disease) in the test set. Using AI with PET/CT alone (protocol A), human visual reading (protocol B), a quantitative method that incorporated maximum standardized uptake value (SUVmax) (protocol C), and a combination of PET/CT, SUVmax, gender, and age (protocol D), obtained data were subjected to ROC curve analyses.

Conclusions: Deep learning with 3D DCNN in combination with FDG-PET/CT imaging results as well as clinical features comprise a novel potential tool shows flexibility for differential diagnosis of MPM.

Copyright: © 2021 Kitajima et al.

In plain English

This retrospective study tested whether an artificial intelligence (AI) method using three-dimensional deep learning could tell malignant pleural mesothelioma (MPM) from benign pleural disease on FDG-PET/CT scans. The authors studied 875 patients examined from 2007 to 2017: 520 with MPM and 355 with benign pleural disease. They compared four approaches on 176 test patients (104 with MPM, 72 benign). These were AI on scans alone (A), visual reading by one experienced reader (B), a scan-uptake measure, SUVmax (C), and AI combining scans with SUVmax, gender and age (D). Areas under the curve were 0.825 (95% confidence interval 0.728 to 0.892), 0.854 (0.775 to 0.908), 0.881 (0.802 to 0.931) and 0.896 (0.818 to 0.943) for A to D. Protocol D performed significantly better than A, B and C. Its sensitivity was 88.5% (92/104; 82.3% to 94.6%) and its specificity 73.6% (53/72; 63.4% to 83.8%). A did not differ significantly from B or from C. The authors called the combined approach a novel, flexible and potentially useful tool. Stated limitations included a retrospective single-institution design, varied scanners and one reader.

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 diagnostic accuracy study comparing AI deep learning, human visual reading and an SUVmax cut-off on FDG-PET/CT scans
• Who was studied: 875 consecutive patients with histologically proven or suspected malignant pleural mesothelioma (751 males, 124 females; mean age 69.1 years, range 28 to 92) examined at the authors' institution from January 2007 to December 2017. 520 had MPM and 355 had benign pleural disease. The cases were split into training (525), validation (174) and test (176) sets.
• Exposure: This is not an asbestos exposure study: the article reports no fiber types or exposure histories. It studied diagnostic method. Four protocols were compared on FDG-PET/CT: AI with imaging alone (A), human visual reading (B), the quantitative measure SUVmax (C), and AI combining imaging, SUVmax, gender and age (D).
• Main results:
  • The test set contained 176 patients: 104 with MPM and 72 with benign pleural disease.
  • In the test set, the area under the ROC curve was 0.825 for protocol A, 0.854 for B, 0.881 for C and 0.896 for D.
  • The 95% confidence intervals for the AUC (Table 2) were 0.728 to 0.892 for A, 0.775 to 0.908 for B, 0.802 to 0.931 for C, and 0.818 to 0.943 for D.
  • Protocol D had significantly better diagnostic performance than A (p = 0.031), B (p = 0.0020) and C (p = 0.041).
  • Protocols B and C differed significantly (p = 0.026). There was no significant difference between A and B (p = 0.38) or between A and C (p = 0.086).
  • Protocol D (AI with imaging, SUVmax, gender and age; cut-off 0.438) had sensitivity 88.5% (92/104; 95% CI 82.3 to 94.6%), specificity 73.6% (53/72; 95% CI 63.4 to 83.8%), and accuracy 82.4% (145/176; 95% CI 76.8 to 88.0%).
  • Protocol A (AI with imaging alone) had sensitivity 77.9% (81/104), specificity 76.4% (55/72), and accuracy 77.3% (136/176).
  • Protocol B (human visual reading) had sensitivity 80.8% (84/104), specificity 77.8% (56/72), and accuracy 79.5% (140/176).
  • Protocol C (SUVmax cut-off 2.23) had sensitivity 85.6% (89/104), specificity 75.0% (54/72), and accuracy 81.3% (143/176).
  • Protocol D's sensitivity was significantly higher than that of A (p = 0.0026) and B (p = 0.013). Its accuracy was significantly higher than that of A (p = 0.0077).
  • In the test set, mean SUVmax was significantly higher for MPM (5.89 plus or minus 3.96) than for benign pleural disease (0.98 plus or minus 1.92), p less than 0.0001.
• The authors' conclusion: The authors concluded that 3D DCNN deep learning combining FDG-PET/CT imaging with clinical features is a novel, flexible and potentially useful tool for the differential diagnosis of MPM.
• Limitations the authors state:
  • The retrospective, single-institution design may limit how far the findings generalize, and statistical errors may be inevitable. The authors said a prospective multicenter study is needed.
  • Different scanners were used over the study period, with differing reconstruction algorithms and acquisition parameters.
  • A single reader did the visual evaluation, rather than two readers reaching a consensus.
  • The AI's indication of which image regions drove its decisions was not precise and not always correct.

Topics

Medical terms in this study, explained in our Lexicon: Pleural Mesothelioma, Flexibility, Differential Diagnosis.

Related studies

• Evaluation and Exploration of Machine Learning and Convolutional Neural Network Classifiers in Detection of Lung Cancer from Microarray Gene-A Paradigm Shift — K. M S, H. Rajaguru, et al. (2023)
• Convolutional Neural Networks for Segmentation of Malignant Pleural Mesothelioma: Analysis of Probability Map Thresholds (CALGB 30901, Alliance) — M. Shenouda, E. Gudmundsson, et al. (2023)
• Diagnosis of mesothelioma with deep learning — X. Hu, Z. Yu (2019)
• Combining a gravitational search algorithm, particle swarm optimization, and fuzzy rules to improve the classification performance of a feed-forward neural network — M.L. Huang, Y.C. Chou (2019)
• Deep learning-based classification of mesothelioma improves prediction of patient outcome — P. Courtiol, C. Maussion, et al. (2019)
• Fully automated volumetric measurement of malignant pleural mesothelioma by deep learning AI: validation and comparison with modified RECIST response criteria — A.C. Kidd, O. Anderson, et al. (2022)

Browse more research in Oncotarget, or return to medical research.