Unsupervised abnormality detection through mixed structure regularization (MSR) in deep sparse autoencoders

Abstract

Purpose The purpose of this study is to introduce and evaluate the mixed structure regularization (MSR) approach for a deep sparse autoencoder aimed at unsupervised abnormality detection in medical images. Unsupervised abnormality detection based on identifying outliers using deep sparse autoencoders is a very appealing approach for computer‐aided detection systems as it requires only healthy data for training rather than expert annotated abnormality. However, regularization is required to avoid overfitting of the network to the training data. Methods We used coronary computed tomography angiography (CCTA) datasets of 90 subjects with expert annotated centerlines. We segmented coronary lumen and wall using an automatic algorithm with manual corrections where required. We defined normal coronary cross section as cross sections with a ratio between lumen and wall areas larger than 0.8. We divided the datasets into training, validation, and testing groups in a tenfold cross‐validation scheme. We trained a deep sparse overcomplete autoencoder model for normality modeling with random structure and noise augmentation. We assessed the performance of our deep sparse autoencoder with MSR without denoising (SAE‐MSR) and with denoising (SDAE‐MSR) in comparison to deep sparse autoencoder (SAE), and deep sparse denoising autoencoder (SDAE) models in the task of detecting coronary artery disease from CCTA data on the test group. Results The SDAE‐MSR achieved the best aggregated area under the curve (AUC) with a 20% improvement and the best aggregated Average Precision (AP) with a 30% improvement upon the SAE and SDAE (AUC: 0.78 to 0.94, AP: 0.66 to 0.86) in distinguishing between coronary cross sections with mild stenosis (stenosis grade < 0.3) and coronary cross sections with severe stenosis (stenosis grade > 0.7). The improvements were statistically significant (Mann–Whitney U‐test, P < 0.001). Similarly, The SDAE‐MSR achieved the best aggregated AUC (AP) with an 18% (18%) improvement upon the SAE and SDAE (AUC: 0.71 to 0.84, AP: 0.68 to 0.80). The improvements were statistically significant (Mann–Whitney U‐test, P < 0.05). Conclusion Deep sparse autoencoders with MSR in addition to explicit sparsity regularization term and stochastic corruption of the input data with Gaussian noise have the potential to improve unsupervised abnormality detection using deep‐learning compared to common deep autoencoders.

Detection of 3D Arterial Centerline Extraction in Spiral CT Coronary Angiography

“…Among these three methods, the manual extraction of the centerline is undoubtedly a simple and reliable method, but when a large amount of data is encountered, the manual extraction method will consume a lot of human and material resources, so it is especially important to study the fully automatic and semi-automatic extraction algorithms [ 14 ]. The fully automatic method generally refers to the process of extracting the centerline without any human intervention, and the algorithm automatically extracts the centerline of the coronary artery based on the input data information; the more classical methods include the method proposed by Freiman to further extract the centerline based on the extraction of the coronary vascular tree by fitting the cylindrical model; the semi-automatic method refers to the process of vessel extraction [ 15 ]. The semi-automatic method refers to the method in which one or several seed points need to be artificially specified as reference points for centerline extraction, and the algorithm extracts the centerline in the image data based on the artificially provided reference points, such as the method proposed by Kojima et al to extract the centerline of coronary arteries after segmenting the aorta and coronary arteries and performing 3D reconstruction by using the local gray values of the vessels and the orientation of the vessels to select the starting point of the iteration [ 16 ].…”

NPBDREG: Uncertainty Assessment in Diffeomorphic Brain MRI Registration using a Non-parametric Bayesian Deep-Learning Based Approach

“…We analyzed the improvement in DNN-based registration generalization capabilities achieved by our NPBDREG approach. We corrupted the input images with two types of noise: Gaussian noise with various std (σ) and mixed structures, which generated a linear combination of the test example (i) and another example sampled randomly from the test set (j): αIj + (1 − α)I i [14]. We used the Dice score and the same labels as in the previous experiment to determine the added-value of our NPB-DREG approach in increasing generalization capabilities of the registration system.…”

Deep autoencoder for false positive reduction in handgun detection

“…Anomaly detection has a wide range of applications. For example, it can be used to detect anomalies in stock prices and time series [21,22], abnormal medical images of findings [23][24][25], abnormal events in video [5,26,27], intrusion detection [28], or disaster areas from radar images [29].…”