Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 Jul 22;11(7):739.
doi: 10.3390/bioengineering11070739.

Automatic Annotation Diagnostic Framework for Nasopharyngeal Carcinoma via Pathology-Fidelity GAN and Prior-Driven Classification

Siqi Zeng  1 , Xinwei Li  2 , Yiqing Liu  3 , Qiang Huang  4 , Yonghong He  3
Affiliations

Automatic Annotation Diagnostic Framework for Nasopharyngeal Carcinoma via Pathology-Fidelity GAN and Prior-Driven Classification

Siqi Zeng et al. Bioengineering (Basel). .

Abstract

Non-keratinizing carcinoma is the most common subtype of nasopharyngeal carcinoma (NPC). Its poorly differentiated tumor cells and complex microenvironment present challenges to pathological diagnosis. AI-based pathological models have demonstrated potential in diagnosing NPC, but the reliance on costly manual annotation hinders development. To address the challenges, this paper proposes a deep learning-based framework for diagnosing NPC without manual annotation. The framework includes a novel unpaired generative network and a prior-driven image classification system. With pathology-fidelity constraints, the generative network achieves accurate digital staining from H&E to EBER images. The classification system leverages staining specificity and pathological prior knowledge to annotate training data automatically and to classify images for NPC diagnosis. This work used 232 cases for study. The experimental results show that the classification system reached a 99.59% accuracy in classifying EBER images, which closely matched the diagnostic results of pathologists. Utilizing PF-GAN as the backbone of the framework, the system attained a specificity of 0.8826 in generating EBER images, markedly outperforming that of other GANs (0.6137, 0.5815). Furthermore, the F1-Score of the framework for patch level diagnosis was 0.9143, exceeding those of fully supervised models (0.9103, 0.8777). To further validate its clinical efficacy, the framework was compared with experienced pathologists at the WSI level, showing comparable NPC diagnosis performance. This low-cost and precise diagnostic framework optimizes the early pathological diagnosis method for NPC and provides an innovative strategic direction for AI-based cancer diagnosis.

Keywords: deep learning; digital staining; nasopharyngeal carcinoma; pathological diagnosis framework; pathology–fidelity.

PubMed Disclaimer

Conflict of interest statement

Qiang Huang is currently employed by Shenzhen Shengqiang Technology Co., Ltd. and was employed by Shenzhen Shengqiang Technology Co., Ltd. while contributing to this manuscript. His contributions to this work and manuscript were made independently without any requirement, guidance or input by my employer. He received no financial compensation from any source for the contributions he made to this scientific work and manuscript.

Figures

Figure A1
Figure A1
Confusion Matrix of Prior-Driven Image Classification System.
Figure 1
Figure 1
Examples of EBER and H&E images. (a,b) are examples of EBER staining, with blue representing NPC tumor cells and pink denoting non-NPC tumor cells (20× magnification). (c,d) are examples of lymphocytic infiltration into tumor cells in H&E images (80× magnification).
Figure 2
Figure 2
Overview of the workflow of PFPD. In PFPD, H&E patches are first cropped from the H&E WSI, and then a GAN model (PF-GAN) is used to achieve digital staining from H&E patches to EBER patches. These digital-stained patches are then input into a prior-driven image classification system (PD-CS) that applies pathological prior knowledge to guide the diagnosis of NPC (0 for EBER-negative, 1 for EBER-positive). Finally, PFPD outputs the WSI-level diagnosis result, a digital-stained EBER WSI and a mask of the tumor regions for the H&E WSI.
Figure 3
Figure 3
The architecture of PF-GAN. PF-GAN introduces the pathology–fidelity constraints by training the generator’s encoder to classify the target domain (0 for EBER-negative, 1 for EBER-positive), extracting global pathological features (such as tumor cell color and morphology). The constraints improve the generator’s ability to achieve accurate pathological mapping when aligning features across different domains, resulting in high-quality images with pathological consistency.
Figure 4
Figure 4
The workflow of the PD-CS. The system consists of three core decision modules: pixel-level, patch-level, and WSI-level. These modules interact via a pixel-to-WSI pathological prior knowledge to guide the diagnostic process.
Figure 5
Figure 5
Visualization of RGB channels’ separation for positive and negative EBER images. The most significant difference between positive and negative is observed in the R channel, followed by the G and B channels.
Figure 6
Figure 6
Histogram of mean pixel intensity distribution for R, G, and B Channels. R channel displays the most considerable threshold range difference between EBER-positive and EBER-negative.
Figure 7
Figure 7
The process of automatic annotation. For PF-GAN training, EBER images are input into the PD-CS system, where annotation is accomplished using its pixel and patch modules. The labeled EBER images and unpaired H&E images are used together as training data for PF-GAN, enabling fully automatic annotation without manual intervention.
Figure 8
Figure 8
Visualization comparison of EBER staining by different GANs at the WSI level. Specifically, Example 1 is a fully negative case, Example 2 presents a case that is easily diagnosed by H&E staining. In such simple cases, PF-GAN accurately transforms large regions of tumor tissue to EBER-positive, while ensuring that small regions of non-NPC tissues remain EBER-negative (as indicated by the red arrows). This highlights PF-GAN’s remarkable capability in preserving the pathology–fidelity.
Figure 9
Figure 9
Visualization comparison of EBER staining by different GANs in organizational regions. The organizational regions shown in the figure are a 1024 × 1024 size block (20× magnification).
Figure 10
Figure 10
Performance of PF-GAN in difficult cases. Both Case 1 and Case 2 exhibit complex tumor microenvironments, with tumor cells infiltrated by lymphocytes.
Figure 11
Figure 11
t-SNE visualization of feature separation for different GANs.
Figure 12
Figure 12
Confusion matrix of diagnostic performance.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Su Z.Y., Siak P.Y., Leong C.O., Cheah S.C. The role of Epstein–Barr virus in nasopharyngeal carcinoma. Front. Microbiol. 2023;14:1116143. doi: 10.3389/fmicb.2023.1116143. - DOI - PMC - PubMed
    1. Wang C.K., Wang T.W., Yang Y.X., Wu Y.T. Deep Learning for Nasopharyngeal Carcinoma Segmentation in Magnetic Resonance Imaging: A Systematic Review and Meta-analysis. Bioengineering. 2024;11:504. doi: 10.3390/bioengineering11050504. - DOI - PMC - PubMed
    1. Jen C.W., Tsai Y.C., Wu J.S., Chen P.L., Yen J.H., Chuang W.K., Cheng S.H. Prognostic classification for patients with nasopharyngeal carcinoma based on American Joint Committee on Cancer staging system T and N categories. Ther. Radiol. Oncol. 2020;4:2. doi: 10.21037/tro.2020.02.01. - DOI
    1. Yoshizaki T., Kondo S., Dochi H., Kobayashi E., Mizokami H., Komura S., Endo K. Recent Advances in Assessing the Clinical Implications of Epstein-Barr Virus Infection and Their Application to the Diagnosis and Treatment of Nasopharyngeal Carcinoma. Microorganisms. 2023;12:14. doi: 10.3390/microorganisms12010014. - DOI - PMC - PubMed
    1. Yang X., Wu J., Chen X. Application of Artificial Intelligence to the Diagnosis and Therapy of Nasopharyngeal Carcinoma. J. Clin. Med. 2023;12:3077. doi: 10.3390/jcm12093077. - DOI - PMC - PubMed

LinkOut - more resources