MesoGraph: Automatic profiling of mesothelioma subtypes from histological images

Our plain-English summary of MesoGraph: Automatic profiling of mesothelioma subtypes from histological images, by M. Eastwood, H. Sailem, S.T. Marc, et al (Cell Reports. Medicine, 2023).

The study

• Authors: M. Eastwood, H. Sailem, S.T. Marc, X. Gao, J. Offman, E. Karteris, A.M. Fernandez, D. Jonigk, W. Cookson, M. Moffatt, S. Popat, F. Minhas, J.L. Robertus
• Journal: Cell Reports. Medicine
• Published: 2023;4(10):101226
• Cited by: 11 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Mesothelioma is classified into three histological subtypes, epithelioid, sarcomatoid, and biphasic, according to the relative proportions of epithelioid and sarcomatoid tumor cells present. Current guidelines recommend that the sarcomatoid component of each mesothelioma is quantified, as a higher percentage of sarcomatoid pattern in biphasic mesothelioma shows poorer prognosis. In this work, we develop a dual-task graph neural network (GNN) architecture with ranking loss to learn a model capable of scoring regions of tissue down to cellular resolution. This allows quantitative profiling of a tumor sample according to the aggregate sarcomatoid association score. Tissue is represented by a cell graph with both cell-level morphological and regional features. We use an external multicentric test set from Mesobank, on which we demonstrate the predictive performance of our model. We additionally validate our model predictions through an analysis of the typical morphological features of cells according to their predicted score.

Copyright © 2023 The Authors. Published by Elsevier Inc. All rights reserved.

In plain English

This study built and tested a computer model, called MesoGraph. It reads microscope images of stained mesothelioma tissue and scores each cell by how closely it is associated with the sarcomatoid or the epithelioid subtype. The study did not examine asbestos exposure. After damaged cores were removed, the model was trained on 234 tissue cores from 90 patients at St. George's Hospital, London, and tested on 258 cores from 77 patients in Mesobank, a UK biobank drawing on several hospitals. To measure how well the model separated subtypes, sarcomatoid and biphasic samples were counted together against epithelioid samples. In internal testing on the St. George's samples, the model's AUROC, a measure of how well it separated the groups, was 0.90. On the Mesobank samples it was 0.86. Patients the model scored as more sarcomatoid had significantly shorter median survival than those scored lower (190 vs. 402 days). After adjustment for gender and age at diagnosis, their hazard ratio was 2.43 (95% CI 1.44 to 4.12). Sarcomatoid tumors tended to have larger and more varied nuclei. The authors concluded the method may offer clinical implications for patient care. They noted limits in interpreting the model and in using small tissue cores.

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: Development and validation of a machine-learning (graph neural network) model for classifying mesothelioma subtypes from tissue images, using two patient cohorts obtained retrospectively
• Who was studied: Malignant mesothelioma patients. The training cohort was 102 patients from St. George's Hospital, London. After dropped and damaged cores were removed, 234 cores from 90 patients were used (148 epithelioid, 61 biphasic, 25 sarcomatoid). The validation cohort was 82 patients from Mesobank, a UK biobank collecting from multiple hospitals. After removals, 258 cores from 77 patients were used (155 epithelioid, 68 biphasic, 35 sarcomatoid).
• Exposure: No asbestos or other exposure was studied. The material analyzed was H&E-stained tissue microarray images of tumor biopsies.
• Main results:
  • For the ROC performance measures, sarcomatoid and biphasic samples were treated together as the positive class and epithelioid samples as the negative class. The model itself was trained with a dual-task ranking approach that ordered sarcomatoid, biphasic and epithelioid cores.
  • In internal cross-validation on the St. George's dataset, MesoGraph's AUROC was 0.90, which was higher than the other methods compared
  • On the external Mesobank validation set, the AUROC was 0.86 and the average precision was 0.8
  • Median survival in the group of patients the model predicted as more sarcomatoid was 190 days, compared with 402 days in the lower-scoring group. The authors described this as significantly shorter.
  • In a Cox model adjusted for gender and age at diagnosis, the hazard ratio for patients the model classed as sarcomatoid (high MesoScore) was 2.43 (95% CI 1.44 to 4.12)
  • Sarcomatoid tumors tended to have larger nuclei on average than epithelioid tumors
  • Sarcomatoid tumors showed higher variability in all nuclear shape features
  • Cell circularity was the feature that contributed most to the model's predictions
  • Biphasic cores were not very distinct from sarcomatoid cores
• The authors' conclusion: The authors concluded that their method characterizes epithelioid and sarcomatoid cells in a quantifiable, reproducible way and may potentially offer clinical implications for patient care
• Limitations the authors state:
  • The model's outputs are hard to interpret: not all of the features it uses correspond to anything a pathologist can see in tissue
  • The model was built on small tissue microarray cores, which are highly selective and contain little surrounding tissue. The authors state that a pipeline trained on whole-slide images would be preferable.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Lung cancer, Mesothelioma.

Medical terms in this study, explained in our Lexicon: Mesothelioma, Histological Subtype, Epithelioid, Sarcomatoid, Biphasic Mesothelioma, Prognosis.

Related studies

• Accuracy of diagnostic biopsy for the histological subtype of malignant pleural mesothelioma — S.C. Kao, T.D. Yan, et al. (2011)
• Two Different Serum MiRNA Signatures Correlate with the Clinical Outcome and Histological Subtype in Pleural Malignant Mesothelioma Patients — M. Lamberti, R. Capasso, et al. (2015)
• Occupational exposure to diesel motor exhaust and risk of lung cancer by histological subtype: a population-based case-control study in Swedish men — A. Ilar, N. Plato, et al. (2017)
• Asbestos exposure and histological subtype of malignant mesothelioma — P. Franklin, H. Alfonso, et al. (2016)
• The Association between the Histological Subtypes of Mesothelioma and Asbestos Exposure Characteristics — T. Vorster, J. Mthombeni, et al. (2022)
• Mesothelioma subtypes of the tunica vaginalis: a rare case report and review of histological criteria — R. Rebecca, M.J. Connor, et al. (2019)

Browse more research in Cell Reports. Medicine, or return to medical research.