See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/346070063

Abnormality Detection in Chest X-Ray Images Using Uncertainty Prediction

Autoencoders

Chapter · September 2020

DOI: 10.1007/978-3-030-59725-2_51

CITATIONS READS

4 231

6 authors, including:

Ruixuan Wang Jianguo Zhang

Sun Yat-Sen University University of Dundee
68 PUBLICATIONS   1,123 CITATIONS    89 PUBLICATIONS   6,808 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Catch a killer - Using the eye to detect and prevent major adverse cardiac events View project

ACE-LP: Augmenting Communication using Environmental data for Language Prediction View project

All content following this page was uploaded by Ruixuan Wang on 12 June 2021.

The user has requested enhancement of the downloaded file.

 Abnormality Detection in Chest X-Ray
Images Using Uncertainty Prediction
Autoencoders

Yifan Mao1 , Fei-Fei Xue1 , Ruixuan Wang1,2(B) , Jianguo Zhang3(B) ,

Wei-Shi Zheng1,2,4 , and Hongmei Liu1,5
1
School of Data and Computer Science, Sun Yat-sen University, Guangzhou, China
[email protected]
2
Key Laboratory of Machine Intelligence and Advanced Computing, MOE,
Guangzhou, China
3
Department of Computer Science and Engineering,
Southern University of Science and Technology, Shenzhen, China
[email protected]
4
Pazhou Lab, Guangzhou, China
5
Guangdong Province Key Laboratory of Information Security Technology,
Guangzhou, China

Abstract. Chest radiography is widely used in annual medical screening

to check whether lungs are healthy or not. Therefore it would be desirable
to develop an intelligent system to help clinicians automatically detect
potential abnormalities in chest X-ray images. Here with only healthy
X-ray images, we propose a new abnormality detection approach based
on an autoencoder which outputs not only the reconstructed normal
version of the input image but also a pixel-wise uncertainty prediction.
Higher uncertainty often appears at normal region boundaries with rela-
tively larger reconstruction errors, but not at potential abnormal regions
in the lung area. Therefore the normalized reconstruction error by the
uncertainty provides a natural measurement for abnormality detection
in images. Experiments on two chest X-ray datasets show the state-of-
the-art performance by the proposed approach.

Keywords: Abnormality detection · Uncertainty prediction ·

Anomaly detection · Chest X-ray

1 Introduction
Chest X-ray has been widely adopted for annual medical screening, where the
main purpose is to check whether the lung is healthy or not. Considering the huge
amount of regular medical tests worldwide, it would be desirable if there exists an
intelligent system helping clinicians automatically detect potential abnormality
in chest X-ray images. Here we consider such a specific task of abnormality detec-
tion, for which there is only normal (i.e., healthy) data available during model
c Springer Nature Switzerland AG 2020
A. L. Martel et al. (Eds.): MICCAI 2020, LNCS 12266, pp. 529–538, 2020.
https://doi.org/10.1007/978-3-030-59725-2_51
530 Y. Mao et al.

training. Compared to diagnosis with supervised learning, the key challenge of

the task is the lack of abnormal data for training an abnormality detector.
For medical image analysis, the approaches thus far proposed for abnormal-
ity detection include parametric and non-parametric statistical models, one-class
SVM, and deep learning models like generative adversarial networks (GANs).
Parametric models usually refer to Gaussian and Gaussian mixture models,
which estimate the density distribution of normal data from training set to pre-
dict the abnormality of a test sample [17]. Parametric models often assume that
the normal data distribution is a Gaussian or a mixture of Gaussian distributions.
In comparison, non-parametric statistical models, such as Gaussian process, are
more capable of modelling complex distributions but have more computational
loads [20]. Both parametric and non-parametric models are bottom-up genera-
tive approaches. In contrast, one-class SVM is a top-down classification-based
method for abnormality detection, which constructs a hyperplane as a decision
boundary that best separates normal data and the origin point, and meanwhile
maximises the distance between the origin and the hyperplane [15]. It has been
applied to abnormal detection based on fMRI and retinal OCT images [11,16].
While the above conventional approaches have been widely used in the med-
ical domain, there is one serious drawback to restrict their performance, i.e.,
the feature representation of images need to be manually designed in advance.
Without the need to extract hand-crafted features, generative adversarial net-
works (GANs) [8] and autoencoders are recently becoming popular for medical
abnormality detection due to their capability of implicitly modelling more com-
plex data distribution than the conventional approaches. The early GAN-based
approach for anomaly detection, called AnoGAN, was proposed for abnormal-
ity detection in retinal OCT images [14]. The basic idea is to train a generator
in the AnoGAN which can generate only normal image patches, such that any
abnormal patch would not be well reconstructed by the generator. A fast version
of the AnoGAN called f-AnoGAN [13] was recently proposed with an additional
encoder included to make the generator become an autoencoder. More autoen-
coder models which are often combined with GANs have also been recently
developed for abnormality detection in medical image analysis [1–4,18] and nat-
ural image analysis [7,12]. One issue in most GAN and autoencoder models is
about the relative large reconstruction errors particularly at region boundaries
although the regions are normal, which would cause false detection of abnormal-
ity in normal images.
This paper for the first time applies an autoencoder model to not only recon-
struct the corresponding normal version of any input image, but also estimate
the uncertainty of reconstruction at each pixel [5,6] to enhance the performance
of anomaly detection. Higher uncertainty often appears at normal region bound-
aries with relatively larger reconstruction errors, but not at potential abnormal
regions in the lung area. As a result, the normalized reconstruction error by
the uncertainty can then be used to better detect potential abnormality. Our
approach obtains state-of-the-art performance on two chest X-ray datasets.
 Abnormality Detection 531

2 Method
The problem of interest is to automatically determine whether any new chest
X-ray image is abnormal (‘unhealthy’) or not, only based on a collection of nor-
mal (‘healthy’) images. Since abnormality in X-ray images could be due to small
area of lesions or unexpected change in subtle contrast between local regions,
extracting an image-level feature representation may suppress such small-scale
features, while extracting features for each local image patch may fail to detect
the contrast-based abnormalities, both resulting in the failing of abnormality
detection. In comparison, reconstruction error based on pixel-level differences
between the original image and its reconstructed version by an autoencoder
model may be a more appropriate measure to detection abnormality in X-ray
images, because both local and global features have been implicitly considered to
reconstruct each pixel by the autoencoder. However, it has been observed that
there often exists relatively large reconstruction errors around the boundaries
between different regions (e.g., lung vs. the others, foreground vs. background,
Fig. 2) even in normal images. Such large errors could result in false positive
detection, i.e., considering a normal image as abnormal. Therefore, it would
be desirable to automatically suppress the contribution of such reconstruction
errors in anomaly detection. Simply detecting edges and removing their contribu-
tions in reconstruction error may not work well due to the difficulty in detecting
low-contrast boundaries in X-ray images and due to possibly larger reconstruc-
tion errors close to region boundaries. In this paper, we applied a probabilistic
approach to automatically downgrade the contribution of normal regions with
larger reconstruction errors. The basic idea is to train an autoencoder to simul-
taneously reconstruct the input image and estimate the pixel-wise uncertainty in
reconstruction (Fig. 1), where larger uncertainties often appear at normal regions
with larger reconstruction errors. On the other hand, there are often relatively
large reconstruction errors with small reconstruction uncertainties at abnormal
regions in the lung area. All together, normal images would be more easily sep-
arated from abnormal images based on the uncertainty-weighted reconstruction
errors.

Fig. 1. Autoencoder with both reconstruction μ(x) and predicted pixel-wise uncer-
tainty σ 2 (x) as outputs.
532 Y. Mao et al.

2.1 Autoencoder with Pixel-Wise Uncertainty Prediction

In order to reconstruct the input image and estimate pixel-wise uncertainty for
the reconstruction, the autoencoder needs to somehow automatically learn to
find where the reconstruction is more uncertain without ground-truth uncer-
tainty available. As in the related work for estimation of uncertainty [5,6,9,10],
here we formulate the reconstruction uncertainty prediction problem by a prob-
abilistic model, with the special (unusual) property that each variance element
in the model is not fixed but varies depending on input data. Formally, given a
collection of N normal images {xi , i = 1, . . . , N }, where xi ∈ RD is the vector-
ized representation of the corresponding i-th original image, an autoencoder can
be trained to make each reconstructed image μ(xi ) as similar to the correspond-
ing input image xi as possible. In general, there are always more or less pixel-
wise differences between the autoencoder’s expected output yi (i.e., same as the
input xi ) and the real output μ(xi ). Suppose such differences are noise sampled
from an input-dependent (note traditionally noise is assumed input-independent)
multivariate Gaussian distribution N (0, Σ(xi )), i.e., yi = μ(xi ) + (xi ), where
(xi ) ∼ N (0, Σ(xi )). Then the conditional probability density of the ideal out-
put yi (same as the input xi ) given the input to the autoencoder is
 
1 1 T −1
p(yi |xi , θ) = D 1 exp − (yi − μ(xi )) Σ (xi )(yi − μ(xi )) ,
(2π) 2 |Σ(xi )| 2 2
(1)
where θ denotes the parameters of the model which can output both the recon-
structed image μ(xi ) and the covariance matrix Σ(xi ). By simplfying Σ(xi ) to
a diagnonal matrix Σ(xi ) = diag(σ12 (xi ), σ22 (xi ), ..., σD2
(xi )), the negative loga-
rithm of Eq. (1) gives
D  
1  (xi,k − μk (xi ))2 D
− log p(yi |xi , θ) = + log σk (xi ) + log(2π),
2
(2)
D σk2 (xi ) 2
k=1

where xi,k is the k-th element of the expected output yi (i.e., the input xi ),
and μk (xi ) is the k-th element of the real output μ(xi ). Then the autoencoder
can be optimized by maximizing the log-likelihood over all the normal (training)
images, i.e., by minimizing the negative log-likelihood function L(θ),
N D  
1   (xi,k − μk (xi ))2
L(θ) = + log σ 2
k (x i ) . (3)
N D i=1 σk2 (xi )
k=1

Equation (3) would be simplified to the mean squared error (MSE) loss based on
either Mahalanobis distance or Euclidean distance, when the variance elements
σk2 (xi )’s are fixed and not dependent on the input xi or when they are not only
fixed but also equivalent.
Note that for each input image xi , the model generates two outputs,
the reconstruction μ(xi ) and the noise variance σ 2 (xi ) = (σ12 (xi ), σ22 (xi ), ...,
σD2
(xi ))T (Fig. 1). Interestingly, while μ(xi ) is supervised to approach to xi ,
 Abnormality Detection 533

σ(xi ) is totally unsupervised during model training, only based on minimization

of the objective function L(θ). From the definition of the noise variance (above
Eq. (1)), each element σk2 (xi ) of the noise variance represents not the reconstruc-
tion error but the degree of uncertainty for the i-th element of the reconstruction
μ(xi ). This uncertainty is used to naturally normalize the reconstruction error
for the i-th element of the reconstruction (first loss term in Eq. (3)). During
model training, the first loss term discourages the autoencoder from predict-
ing very small uncertainty values for those pixels with higher reconstruction
errors, because smaller σk2 (xi ) will enlarge the contribution of the already large
reconstruction errors by the first loss term. Therefore, the autoencoder will auto-
matically learn to generate relatively larger uncertainties for those pixels (e.g.,
around region boundaries) with relatively larger reconstruction errors in normal
images. On the other hand, the second loss term log σk2 (xi ) in Eq. (3) will prevent
the autoencoder from predicting larger uncertainty for all reconstructed pixels.
Therefore, the two loss terms together will help train an autoencoder such that
the predicted uncertainty will be smaller at those regions where the model can
reconstruct well and relatively larger otherwise in normal images.
It is worth noting that the positive correlation between the uncertainty pre-
diction and the reconstruction error may hold mainly for normal image pix-
els or regions. For anomaly in the lung area which has not been seen during
model training, the uncertainty prediction is often small (see Sect. 3.2), proba-
bly because the model has learned to reconstruct well (with smaller uncertainty)
inside the lung area during model training and therefore often predicts low uncer-
tainty for lung area for any new image, no matter whether there exists anomaly in
the area or not. On the other hand, the reconstruction errors at abnormal regions
in the lung area are often relatively large because the well-trained autoencoder
learns to just reconstruct normal lung by removing any potential noise or abnor-
mal signals in this area. As a result, anomaly with larger reconstruction errors
and small uncertainty would become distinctive from normal regions which have
positive correlation between reconstruction errors and predicted uncertainties.

2.2 Abnormality Detection

Based on the above analysis, for any new image x, it is natural to use the pixel-
wise normalized reconstruction error (as first term in Eq. (3)) to represents the
degree of abnormality for each pixel xk , and the average of such errors over all
pixels for the abnormality A(x) of the image, i.e.,
D
1  (xk − μk (x))2
A(x) = . (4)
D σk2 (x)
k=1

Since the pixel-wise uncertainties σk2 (x) depend on the input x, it is not as easily
estimated as for fixed variance. As far as we know, it is the first time to apply
such pixel-wise input-dependent uncertainty to estimate of abnormality. If the
image x is normal, pixels or regions with larger reconstruction errors are often
534 Y. Mao et al.

accompanied with larger uncertainties, therefore often resulting in the overall

smaller abnormality score A(x). In contrast, if there is certain anomaly in the
image, the relatively larger reconstruction errors still with small uncertainties at
the abnormal region would lead to a relatively larger abnormality score A(x).

3 Experiments
3.1 Experimental Setup
Datasets. Our method is tested on two publicly available chest X-ray datasets:
1) RSNA Pneumonia Detection Challenge dataset1 and 2) pediatric chest X-ray
dataset2 . The RSNA dataset is a subset of ChestXray14 [19]; it contains 26,684
X-rays with 8,851 normal, 11,821 no lung opacity/not normal and 6,012 lung
opacity. The pediatric dataset consists of 5,856 X-rays from normal children and
patients with pneumonia.
Protocol. For the RSNA dataset, we used 6,851 normal images for training,
1,000 normal and 1,000 abnormal images for testing. On this dataset, our method
was tested on three different settings: 1) normal vs. lung opacity; 2)normal
vs. not normal and 3) normal vs. all (lung opacity and not normal). For the
pediatric dataset, 1,249 normal images were used for training, and the original
author-provided test set was used to evaluate the performance. The test set
contains 234 normal images and 390 abnormal images. All images were resized
to 64 × 64 pixels and pixel values of each image were normalized to [-1,1]. The
area under the ROC curve (AUC) is used to evaluate the performance, together
with equal error rate (EER), F1-score (at EER) reported.
Implementation. The backbone of our method is a convolutional autoencoder.
The network is symmetric containing an encoder and a decoder. The encoder
contains four layers (each with one 4 × 4 convolution with a stride 2), which
is then followed by two fully connected layers whose output sizes are 2048 and
16 respectively. The decoder is connected by two fully connected layers and
four transposed convolutions, which constitute the encoder. The channel sizes
are 16-32-64-64 for encoder and 64-64-32-16 for decoder. All convolutions and
transposed convolutions are followed by batch normalization and ReLU nonlin-
earity except for the last output layer. We trained our model for 250 epochs. The
optimization was done using the Adam optimizer with a learning rate 0.0005.
For numerical stability we did not directly predict σ 2 in Eq. (3). Instead, the
uncertainty output by the model is the log variance (i.e., log σ 2 ).

3.2 Evaluations
Baselines. Our method is compared with three baselines as well as state-of-the-
art methods for anomaly detection. Below summarizes the methods compared.
1
https://www.kaggle.com/c/rsna-pneumonia-detection-challenge.
2
https://doi.org/10.17632/rscbjbr9sj.3.
 Abnormality Detection 535

Table 1. Comparison with others with different metrics. Bold face indicates the best,
and italic face for the second best.

Method RSNA Setting-1 RSNA Setting-2 RSNA Setting-3 Pediatric

EER↓ F1↑ AUC↑ EER↓ F1↑ AUC↑ EER↓ F1↑ AUC↑ EER↓ F1↑ AUC↑
AE 0.36 0.64 0.68 0.40 0.60 0.63 0.38 0.62 0.65 0.41 0.65 0.64
OC-SVM-1 0.41 0.59 0.63 0.45 0.54 0.57 0.42 0.58 0.60 0.38 0.67 0.67
OC-SVM-2 0.31 0.69 0.74 0.40 0.60 0.64 0.46 0.64 0.69 0.39 0.66 0.68
f-AnoGAN 0.21 0.79 0.84 0.31 0.68 0.73 0.27 0.73 0.79 0.33 0.72 0.71
Ours 0.18 0.81 0.89 0.28 0.72 0.78 0.22 0.77 0.83 0.29 0.75 0.78

– Autoencoder (AE). A vanilla autoencoder is the most relevant baseline. For a

fair comparison, the backbone of the vanilla AE is designed exactly the same
as ours. We use the L2 reconstruction error as anomaly score for this method.
– OC-SVM. The one-class support vector machine (OC-SVM)[15] is a tradi-
tional model for one-class learning. For OC-SVM, we use the feature repre-
sentations (i.e., the output of the encoder) learned from a vanilla AE and
ours as the input to SVM respectively, resulting in two versions OC-SVM-1
and OC-SVM-2.
– f-AnoGAN. It is a state-of-the-art anomaly detection method in medical imag-
ing [13]. During inference in this model, we fed an image into the encoder-
generator to acquire an reconstructed image. A hybrid score combining pixel-
level and feature reconstruction error is used to measure abnormality.

Comparison and analysis. The abnormality detection performance with

different methods was summarized in Table 1. The state-of-the-art method f-
AnoGAN clearly outperforms the other baselines, but performs worse than ours.
OC-SVM-2 (with our encoder) is consistently better than OC-SVM-1, suggest-
ing that the encoder in our approach may have mapped normal data into a
more compact region in the latent feature space, which can be easily learned
by one-class SVM. The superior performance of our method is probably due to
the suppression of larger reconstruction error at normal region boundaries by
the predicted pixel-wise uncertainties. As Fig. 2 (columns 3, 5, 7) demonstrated,
while the reconstruction errors are relatively large at some normal region bound-
aries for all methods, only our method can estimate the pixel-wise uncertainty
(column 8), by which the pixel-wise normalized reconstruction errors at nor-
mal region boundaries has been largely reduced (column 9). On the other hand,
larger reconstruction errors in abnormal regions in the lung area often do not
correspond to larger uncertainties.
As a result, the uncertainty normalized abnormality score can help separate
abnormal images from normal ones, as confirmed in Fig. 3 (right). In comparison,
the two histograms are largely overlapped when using the vanilla reconstruction
error (Fig. 3, left). In addition, it is worth noting that, as in other autoencoder
and GAN based image reconstruction methods, our method can also provide the
pixel-level localization of potential abnormalities (Fig. 2, last column), which
could be helpful for clinicians to check and analyze the abnormality details in
practice.
536 Y. Mao et al.

Fig. 2. Exemplar reconstructions of normal (rows 1–2) and abnormal (rows 3–4) test
images. x is input; x , x , and μ(x) are reconstructions from AE, f-AnoGAN, and our
method; operators are pixel-wise. Green bounding boxes for abnormal regions.

Fig. 3. Histograms of abnormality score for normal (blue) and abnormal (red) images
in the test set (RSNA Setting-1). Left: without uncertainty normalization. Right: with
uncertainty normalization. Scores are normalized to [0, 1] in each subfigure. (Color
figure online)

Ablation Study. Table 2 shows that only incorporating uncertainty loss with
autoencoder (i.e., without uncertainty normalization) doesn’t improve the per-
formance (Table 2, ‘without-U’, AUC = 0.68 which is similar to that of vanilla
AE). In contrast, uncertainty normalized abnormality score (‘with-U’) largely
improves the performance. Interestingly, adding skip connections downgraded

Table 2. Ablation study on RSNA Setting-1. ‘U’ denotes uncertainty output. ‘0’–‘4’:
number of skip connections between encoder and decoder convolutional layers, with ‘1’
for the connection between encoder’s last and decoder’s first convolutional layers.

Skip connections 0 1 2 3 4
with-U 0.89 0.62 0.50 0.44 0.38
without-U 0.68 0.43 0.38 0.33 0.33
 Abnormality Detection 537

performance. This is probably because skip connections prevents the encoder

learning the true low-dimensional distribution of normal data.

4 Conclusion

We proposed an uncertainty normalized abnormality detection method which is

capable of reconstructing the image with the pixel-wise prediction uncertainty.
Experiments on two chest X-ray datasets shows that the uncertainty can well
suppress the adversarial effect of larger reconstruction errors around normal
region boundaries, and consequently state-of-the-art performance was obtained.

Acknowledgement. This work is supported in part by the National Key Research

and Development Program (grant No. 2018YFC1315402), the Guangdong Key
Research and Development Program (grant No. 2019B020228001), the National Nat-
ural Science Foundation of China (grant No. U1811461), the Guangzhou Science and
Technology Program (grant No. 201904010260) and the National Key R&D Program
of China (grant No. 2017YFB0802500).

References
1. Alaverdyan, Z., Jung, J., Bouet, R., Lartizien, C.: Regularized siamese neural net-
work for unsupervised outlier detection on brain multiparametric magnetic res-
onance imaging: application to epilepsy lesion screening. Med. Image Anal. 60,
101618 (2020)
2. Baur, C., Wiestler, B., Albarqouni, S., Navab, N.: Fusing unsupervised and super-
vised deep learning for white matter lesion segmentation. In: International Confer-
ence on Medical Imaging with Deep Learning, pp. 63–72 (2019)
3. Chen, X., Konukoglu, E.: Unsupervised detection of lesions in brain MRI using
constrained adversarial auto-encoders. In: International Conference on Medical
Imaging with Deep Learning (2018)
4. Chen, X., Pawlowski, N., Glocker, B., Konukoglu, E.: Unsupervised lesion detection
with locally Gaussian approximation. In: Suk, H.-I., Liu, M., Yan, P., Lian, C. (eds.)
MLMI 2019. LNCS, vol. 11861, pp. 355–363. Springer, Cham (2019). https://doi.
org/10.1007/978-3-030-32692-0 41
5. Diederik, P.K., Welling, M., et al.: Auto-encoding variational bayes. In: Proceedings
of the International Conference on Learning Representations, vol. 1 (2014)
6. Dorta, G., Vicente, S., Agapito, L., Campbell, N.D., Simpson, I.: Structured uncer-
tainty prediction networks. In: Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp. 5477–5485 (2018)
7. Gong, D., et al.: Memorizing normality to detect anomaly: memory-augmented
deep autoencoder for unsupervised anomaly detection. In: Proceedings of the IEEE
International Conference on Computer Vision, pp. 1705–1714 (2019)
8. Goodfellow, I., et al.: Generative adversarial nets. In: Advances in Neural Infor-
mation Processing Systems, pp. 2672–2680 (2014)
9. He, Y., Zhu, C., Wang, J., Savvides, M., Zhang, X.: Bounding box regression with
uncertainty for accurate object detection. In: Proceedings of the IEEE Conference
on Computer Vision and Pattern Recognition, pp. 2888–2897 (2019)
538 Y. Mao et al.

10. Kendall, A., Gal, Y.: What uncertainties do we need in Bayesian deep learning
for computer vision? In: Advances in Neural Information Processing Systems, pp.
5574–5584 (2017)
11. Mourão-Miranda, J., et al.: Patient classification as an outlier detection problem:
an application of the one-class support vector machine. NeuroImage 58(3), 793–804
(2011)
12. Sabokrou, M., Khalooei, M., Fathy, M., Adeli, E.: Adversarially learned one-class
classifier for novelty detection. In: Proceedings of the IEEE Conference on Com-
puter Vision and Pattern Recognition, pp. 3379–3388 (2018)
13. Schlegl, T., Seeböck, P., Waldstein, S.M., Langs, G., Schmidt-Erfurth, U.: f-
AnoGAN: fast unsupervised anomaly detection with generative adversarial net-
works. Med. Image Anal. 54, 30–44 (2019)
14. Schlegl, T., Seeböck, P., Waldstein, S.M., Schmidt-Erfurth, U., Langs, G.: Unsu-
pervised anomaly detection with generative adversarial networks to guide marker
discovery. In: Niethammer, M., et al. (eds.) IPMI 2017. LNCS, vol. 10265, pp.
146–157. Springer, Cham (2017). https://doi.org/10.1007/978-3-319-59050-9 12
15. Schölkopf, B., Williamson, R.C., Smola, A.J., Shawe-Taylor, J., Platt, J.C.: Sup-
port vector method for novelty detection. In: Advances in Neural Information Pro-
cessing Systems, pp. 582–588 (2000)
16. Seeböck, P., et al.: Unsupervised identification of disease marker candidates in
retinal OCT imaging data. IEEE Trans. Med. Imaging 38(4), 1037–1047 (2018)
17. Sidibe, D., et al.: An anomaly detection approach for the identification of DME
patients using spectral domain optical coherence tomography images. Comput.
Methods Programs Biomed. 139, 109–117 (2017)
18. Tang, Y.X., Tang, Y.B., Han, M., Xiao, J., Summers, R.M.: Abnormal chest x-ray
identification with generative adversarial one-class classifier. In: IEEE International
Symposium on Biomedical Imaging, pp. 1358–1361 (2019)
19. Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., Summers, R.M.: ChestX-ray8:
hospital-scale chest x-ray database and benchmarks on weakly-supervised classi-
fication and localization of common thorax diseases. In: Proceedings of the IEEE
Conference on Computer Vision and Pattern Recognition, pp. 2097–2106 (2017)
20. Ziegler, G., Ridgway, G.R., Dahnke, R., Gaser, C., Initiative, A.D.N., et al.: Indi-
vidualized Gaussian process-based prediction and detection of local and global gray
matter abnormalities in elderly subjects. NeuroImage 97, 333–348 (2014)

View publication stats