Predicting Semantic Descriptions from Medical Images with Convolutional Neural Networks

• Published in: Information Processing in Medical Imaging Vol. 24; pp. 437 - 448
• Main Authors: Schlegl, Thomas, Waldstein, Sebastian M., Vogl, Wolf-Dieter, Schmidt-Erfurth, Ursula, Langs, Georg
• Format: Book Chapter Journal Article
• Language: English
• Published: Cham Springer International Publishing 2015
• Series: Lecture Notes in Computer Science
• Subjects: Algorithms; Class Label; Convolutional Neural Network; Humans; Image Enhancement - methods; Image Interpretation, Computer-Assisted - methods; Image Patch; Neural Networks (Computer); Object Class; Pattern Recognition, Automated - methods; Reproducibility of Results; Retina - pathology; Retinal Diseases - pathology; Semantic Description; Semantics; Sensitivity and Specificity; Tomography, Optical Coherence - methods
• ISBN: 9783319199917; 3319199919
• ISSN: 0302-9743; 1011-2499; 1611-3349
• DOI: 10.1007/978-3-319-19992-4_34

Learning representative computational models from medical imaging data requires large training data sets. Often, voxel-level annotation is unfeasible for sufficient amounts of data. An alternative to manual annotation, is to use the enormous amount of knowledge encoded in imaging data and corresponding reports generated during clinical routine. Weakly supervised learning approaches can link volume-level labels to image content but suffer from the typical label distributions in medical imaging data where only a small part consists of clinically relevant abnormal structures. In this paper we propose to use a semantic representation of clinical reports as a learning target that is predicted from imaging data by a convolutional neural network. We demonstrate how we can learn accurate voxel-level classifiers based on weak volume-level semantic descriptions on a set of 157 optical coherence tomography (OCT) volumes. We specifically show how semantic information increases classification accuracy for intraretinal cystoid fluid (IRC), subretinal fluid (SRF) and normal retinal tissue, and how the learning algorithm links semantic concepts to image content and geometry.