Deciphering serous ovarian carcinoma histopathology and platinum response by convolutional neural networks - PubMed (original) (raw)

Deciphering serous ovarian carcinoma histopathology and platinum response by convolutional neural networks

Kun-Hsing Yu et al. BMC Med. 2020.

Abstract

Background: Ovarian cancer causes 151,900 deaths per year worldwide. Treatment and prognosis are primarily determined by the histopathologic interpretation in combination with molecular diagnosis. However, the relationship between histopathology patterns and molecular alterations is not fully understood, and it is difficult to predict patients' chemotherapy response using the known clinical and histological variables.

Methods: We analyzed the whole-slide histopathology images, RNA-Seq, and proteomics data from 587 primary serous ovarian adenocarcinoma patients and developed a systematic algorithm to integrate histopathology and functional omics findings and to predict patients' response to platinum-based chemotherapy.

Results: Our convolutional neural networks identified the cancerous regions with areas under the receiver operating characteristic curve (AUCs) > 0.95 and classified tumor grade with AUCs > 0.80. Functional omics analysis revealed that expression levels of proteins participated in innate immune responses and catabolic pathways are associated with tumor grade. Quantitative histopathology analysis successfully stratified patients with different response to platinum-based chemotherapy (P = 0.003).

Conclusions: These results indicated the potential clinical utility of quantitative histopathology evaluation in tumor cell detection and chemotherapy response prediction. The developed algorithm is easily extensible to other tumor types and treatment modalities.

Keywords: Digital pathology; Gene expression; Machine learning; Platinum response; Proteomics; Serous ovarian carcinoma.

PubMed Disclaimer

Conflict of interest statement

To maximize the impact of this study, Harvard Medical School submitted a non-provisional patent application to the United States Patent and Trademark Office (USPTO).

Figures

Fig. 1

Fig. 1

Integrative histopathology-functional omics analyses on serous ovarian carcinoma. a A model of the informatics workflow in this study. b Convolutional neural networks identified regions with tumor cells of serous ovarian carcinoma. Receiver operating characteristic (ROC) curves of convolutional neural networks that classified regions with tumor cells from those without tumor cells in the independent test set are shown. Areas under the receiver operating characteristic curves (AUCs) in the independent test set: AlexNet = 0.955 ± 0.010; GoogLeNet = 0.974 ± 0.004; VGGNet = 0.975 ± 0.001. c Gradient-weighted class activation maps (grad-CAMs) confirmed that the CNN models focused on the cancerous part of the histopathology slides when classifying malignant tissues from benign ones. The original hematoxylin-and-eosin-stained histopathology image was also shown

Fig. 2

Fig. 2

Quantitative histopathology analysis identified tumor grade. a ROC curves of convolutional neural networks that classified the pathology grade of serous ovarian carcinoma. The sensitivity and specificity for identifying high-grade serous ovarian carcinoma are shown. AUC in the independent test set: AlexNet = 0.760 ± 0.082; GoogLeNet = 0.810 ± 0.067; VGGNet = 0.812 ± 0.088. b The gradient-weighted class activation map (grad-CAM) of a histopathology image of a low-grade ovarian cancer patient and the original hematoxylin-and-eosin-stained histopathology image. Tumor cells and differentiated cellular structures received higher weighted in the grad-CAM. c The grad-CAM of a histopathology image of a high-grade ovarian cancer patient and the original hematoxylin-and-eosin-stained histopathology image. Clusters of tumor cells with poor differentiation were highlighted by the grad-CAM

Fig. 3

Fig. 3

Proteomics analyses revealed the molecular profiles associated with tumor grade. a The expression levels of 32 proteins are associated with tumor grade. Sidebar: red indicates high-grade tumors; blue indicates low-to-moderate-grade tumors. b The protein-protein interaction (PPI) network of the proteins associated with tumor grade. These 32 proteins have significantly enriched PPIs (P < 6.66 × 10−16). The color of the edges shows the information source of the curated protein-protein interactions in the STRING database

Fig. 4

Fig. 4

Convolutional neural networks predicted the platinum-based chemotherapy response of patients with serous ovarian carcinoma. a Convolutional neural networks stratified serous ovarian carcinoma patients with different platinum-based chemotherapy response (log-rank test P = 0.003). Kaplan-Meier curves of the image-based stratification in the test cases are shown. b The gradient-weighted class activation map (grad-CAM) of a histopathology image of a serous ovarian carcinoma patient with short platinum-free interval (PFI) and the original hematoxylin-and-eosin-stained histopathology image. c The grad-CAM of a histopathology image of a serous ovarian carcinoma patient with long PFI and the original hematoxylin-and-eosin-stained histopathology image. Grad-CAMs highlighted regions occupied by the tumor cells

Fig. 5

Fig. 5

Proteomic profiles are associated with platinum-based chemotherapy response. a 72 proteins are significantly associated with the platinum-free interval (PFI) of serous ovarian cancer patients. b The interaction network of the proteins associated with PFI. Proteins associated with platinum response have significantly enriched protein-protein interactions (P = 7.86 × 10−9). The color of the edges shows the information source of the curated protein-protein interactions in the STRING database

Similar articles

Cited by

References

    1. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer statistics, 2012. CA Cancer J Clin. 2015;65(2):87–108. doi: 10.3322/caac.21262. - DOI - PubMed
    1. Torre LA, Trabert B, DeSantis CE, Miller KD, Samimi G, Runowicz CD, Gaudet MM, Jemal A, Siegel RL. Ovarian cancer statistics, 2018. CA Cancer J Clin. 2018;68(4):284-96. - PMC - PubMed
    1. Mutch DG, Prat J. 2014 FIGO staging for ovarian, fallopian tube and peritoneal cancer. Gynecol Oncol. 2014;133(3):401–404. doi: 10.1016/j.ygyno.2014.04.013. - DOI - PubMed
    1. Jayson GC, Kohn EC, Kitchener HC, Ledermann JA. Ovarian cancer. Lancet. 2014;384(9951):1376–1388. doi: 10.1016/S0140-6736(13)62146-7. - DOI - PubMed
    1. Kosary CL. FIGO stage, histology, histologic grade, age and race as prognostic factors in determining survival for cancers of the female gynecological system: an analysis of 1973-87 SEER cases of cancers of the endometrium, cervix, ovary, vulva, and vagina. Semin Surg Oncol. 1994;10(1):31–46. doi: 10.1002/ssu.2980100107. - DOI - PubMed

Publication types

MeSH terms

Substances

LinkOut - more resources