arXiv subjects

Ursula Schmidt-Erfurth

Publications and source records attributed to Ursula Schmidt-Erfurth.

39 records · Page 3Linked to original sources

arXiv2017

Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery

Obtaining models that capture imaging markers relevant for disease progression and treatment monitoring is challenging. Models are typically based on large amounts of data with annotated examples of known markers aiming at automating detection. High annotation effort and the limitation to a vocabulary of known markers limit the power of such approaches. Here, we perform unsupervised learning to identify anomalies in imaging data as candidates for markers. We propose AnoGAN, a deep convolutional generative adversarial network to learn a manifold of normal anatomical variability, accompanying a novel anomaly scoring scheme based on the mapping from image space to a latent space. Applied to new data, the model labels anomalies, and scores image patches indicating their fit into the learned distribution. Results on optical coherence tomography images of the retina demonstrate that the approach correctly identifies anomalous images, such as images containing retinal fluid or hyperreflective foci.

cs.CV↗

arXiv2016

Reproducibility of Retinal Thickness Measurements across Spectral-Domain Optical Coherence Tomography Devices using Iowa Reference Algorithm

PURPOSE: Establishing and obtaining consistent quantitative indices of retinal thickness from a variety of clinically used Spectral-Domain Optical Coherence Tomography scanners. DESIGN: Retinal images from five Spectral-Domain Optical Coherence Tomography scanners were used to determine total retinal thickness with scanner-specific correction factors establishing consistency of thickness measurement across devices. PARTICIPANTS: 55 Fovea-centered Spectral-Domain Optical Coherence Tomography volumes from eleven subjects were analyzed, obtained from Cirrus HD-OCT, RS-3000, Heidelberg Spectralis, RTVue and Topcon2000, seven subjects with retinal diseases and four normal controls. METHOD: The Iowa Reference Algorithm measured total retinal thickness. Nonlinear model of total retinal thickness measurement comparisons was derived and used for device-specific comparisons. Bland-Altman plots and pairwise predictive equations yielded pairwise scanner-specific differences. Mendel test determined whether measurement biases were constant for each scanner pair. RESULTS: Mendel test revealed that all pairwise scanner differences of total retinal thickness were constant across the cohort (p=0.992). Individual measurements can thus be bias-corrected and the Iowa Reference Algorithm serve as a scanner-agnostic independent standard of total retinal thickness across the five tested SD-OCT scanners. CONCLUSIONS: Combination of the Iowa Reference Algorithm with scanner-specific bias correction yields cross-scanner consistency of total retinal thickness measurements, facilitating scanning-device independent quantitative assessment of total retinal thickness, longitudinal follow-up quantification without requiring patients to be imaged on the same scanner model, and allowing for multi-center studies with heterogeneous device utilization when using the Iowa Reference Algorithm.

physics.med-ph↗

arXiv2016

Identifying and Categorizing Anomalies in Retinal Imaging Data

The identification and quantification of markers in medical images is critical for diagnosis, prognosis and management of patients in clinical practice. Supervised- or weakly supervised training enables the detection of findings that are known a priori. It does not scale well, and a priori definition limits the vocabulary of markers to known entities reducing the accuracy of diagnosis and prognosis. Here, we propose the identification of anomalies in large-scale medical imaging data using healthy examples as a reference. We detect and categorize candidates for anomaly findings untypical for the observed data. A deep convolutional autoencoder is trained on healthy retinal images. The learned model generates a new feature representation, and the distribution of healthy retinal patches is estimated by a one-class support vector machine. Results demonstrate that we can identify pathologic regions in images without using expert annotations. A subsequent clustering categorizes findings into clinically meaningful classes. In addition the learned features outperform standard embedding approaches in a classification task.

cs.LG↗