Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2019 Oct 16;11(10):1579.
doi: 10.3390/cancers11101579.

Prediction of BAP1 Expression in Uveal Melanoma Using Densely-Connected Deep Classification Networks

Affiliations
Free PMC article

Prediction of BAP1 Expression in Uveal Melanoma Using Densely-Connected Deep Classification Networks

Muyi Sun et al. Cancers (Basel). .
Free PMC article

Abstract

Uveal melanoma is the most common primary intraocular malignancy in adults, with nearly half of all patients eventually developing metastases, which are invariably fatal. Manual assessment of the level of expression of the tumor suppressor BRCA1-associated protein 1 (BAP1) in tumor cell nuclei can identify patients with a high risk of developing metastases, but may suffer from poor reproducibility. In this study, we verified whether artificial intelligence could predict manual assessments of BAP1 expression in 47 enucleated eyes with uveal melanoma, collected from one European and one American referral center. Digitally scanned pathology slides were divided into 8176 patches, each with a size of 256 × 256 pixels. These were in turn divided into a training cohort of 6800 patches and a validation cohort of 1376 patches. A densely-connected classification network based on deep learning was then applied to each patch. This achieved a sensitivity of 97.1%, a specificity of 98.1%, an overall diagnostic accuracy of 97.1%, and an F1-score of 97.8% for the prediction of BAP1 expression in individual high resolution patches, and slightly less with lower resolution. The area under the receiver operating characteristic (ROC) curves of the deep learning model achieved an average of 0.99. On a full tumor level, our network classified all 47 tumors identically with an ophthalmic pathologist. We conclude that this deep learning model provides an accurate and reproducible method for the prediction of BAP1 expression in uveal melanoma.

Keywords: BAP1 expression prediction; artificial intelligence; deep learning; densely-connected network; immunohistochemistry; ophthalmic histopathology images; precision medicine.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1
Figure 1
One sample from the dataset of the original scanning of the BAP1 stained uveal melanoma slides and the image patches (256 × 256) cropped from the regions of interest in this image. All patches were divided into four categories: P-positive, N-negative, B-blurred, and E-excluded. We randomly sampled four patches in each category for illustration.
Figure 2
Figure 2
The data flow in our research. First, the raw images were cropped into patches. Second, the patches were finely annotated through two steps by an ophthalmic pathologist. Finally, the dataset was separated into two subsets and fed into the network for training and prediction.
Figure 3
Figure 3
Illustration of our densely-connected deep network. We employed the DenseNet-121 based network in which there are four dense blocks. These four dense blocks are composed by densely-connected cascaded convolutional operations with different groups of convolutions with the number of 6, 12, 24, 16, as shown in the left bottom. The dense connection is shown by five convolutional groups. Each group of convolutions is composed of one 1 × 1 convolutional layer and one 3 × 3 convolutional layer. All of the dense blocks are connected by one 1 × 1 convolutional layer and one 2 × 2 pooling layer.
Figure 4
Figure 4
Receiver operating characteristic (ROC) curves of the four categories. Class positive = retained nuclear BAP1 expression. Class negative = lost nuclear BAP1 expression.
Figure 5
Figure 5
Kaplan-Meier curve. Patients with tumors that had low BAP1-expression, identically classified by both an ophthalmic pathologist and the deep learning network, had significantly shorter cumulative metastasis-free survival than patients with tumors that had high BAP1-expression (Log-Rank p = 0.000009).
Figure 6
Figure 6
Several samples of convolutional feature maps in our networks with different downsampling strides. We visualized two groups of feature maps from two categories: positive and negative. With the increase of depth, the feature maps became more abstract from detailed enhancement to global perception. For different categories, we sampled the feature maps with the same layers and channels (four channels in each layer). The coding of feature maps from the two categories was distinct.
Figure 7
Figure 7
Visualization of the effectiveness of our network. (Line 1) Three raw histopathology images with annotations. (Line 2) The corresponding predictions in the regions of interest of the three samples. Yellow, green, red, and blue corresponds to positive (retained nuclear BAP1 expression), negative (lost nuclear BAP1 expression), excluded, and blurred, respectively.
Figure 8
Figure 8
Prediction of BAP1 classification with our network in one detailed area of interest. (Left) Overview of the specimen with a “collar button” configuration. (Middle left) Original image of the region outlined in the blue box. (Middle right) Annotation by ophthalmic pathologist. (Right) Prediction by our network. Yellow areas correspond to BAP1-classification “high” and green to “low”.

Similar articles

See all similar articles

Cited by 2 articles

References

    1. Bray F., Ferlay J., Soerjomataram I., Siegel R.L., Torre L.A., Jemal A. Global cancer statistics 2018: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA-Cancer J. Clin. 2018;68:394–424. doi: 10.3322/caac.21492. - DOI - PubMed
    1. Stang A., Parkin D., Ferlay J., Jöckel K.-H. International uveal melanoma incidence trends in view of a decreasing proportion of morphological verification. Int. J. Cancer. 2005;114:114–123. doi: 10.1002/ijc.20690. - DOI - PubMed
    1. Virgili G., Gatta G., Ciccolallo L., Capocaccia R., Biggeri A., Crocetti E., Lutz J.-M.., Paci E. Survival in patients with uveal melanoma in europe. Arch. Ophthalmol. 2008;126:1413–1418. doi: 10.1001/archopht.126.10.1413. - DOI - PubMed
    1. Jovanovic P., Mihajlovic M., Djordjevic-Jocic J., Vlajkovic S., Cekic S., Stefanovic V. Ocular melanoma: An overview of the current status. Int. J. Clin. Exp. Pathol. 2013;6:1230–1244. - PMC - PubMed
    1. Kujala E., Makitie T., Kivela T. Very long-term prognosis of patients with malignant uveal melanoma. Investig. Ophthalmol. Vis. Sci. 2003;44:4651–4659. doi: 10.1167/iovs.03-0538. - DOI - PubMed
Feedback