Automatic classfication of mammograhic breast density
Cristina Rossi, Alexander Ciritsis, Ilaria De Martini, Matthias Eberhard, Magda Marcon, Anton S. Becker, Nicole Berger, Andreas Boss (USZ, UZH)
High breast density is a risk factor for breast cancer. In addition to its relevance for the assessment of individual breast cancer risk, mammographic density (MD) or breast density is an important parameter in the planning of systematic mammography screening programs. Patients with dense breasts may need additional imaging, such as tomosynthesis, ultrasound or breast MR, to increase the chances of cancer detection. However, the evaluation of MD is strongly reader dependent with low accuracy between readers. The aim of this study was the development of a deep convolutional neural network (dCNN) for the automatic classification of breast density based on the mammographic appearance of the tissue according to the American College of Radiology Breast Imaging and Data System (ACR BI-RADS) Atlas. Tested against expert readers, the dCNN provided accurate classification of breast density based on the ACR BI-RADS system. The proposed technique may allow for an accurate, standardized and observer-independent breast density assessment of mammograms.
Trace and detect adversarial attacks on CNNs using feature response maps
Mohammadreza Amirian, Friedhelm Schwenker, Thilo Stadelmann (ZHAW, University Ulm)
The existence of adversarial attacks on convolutional neural networks (CNN) questions the fitness of such models for serious applications. The attacks manipulate an input image such that misclassification is evoked while still looking normal to a human observer—they are thus not easily detectable. In a different context, backpropagated activations of CNN hidden layers—“feature responses” to a given input—have been helpful to visualize for a human “debugger” what the CNN “looks at” while computing its output. In this work, we propose a novel detection method for adversarial examples to prevent attacks. We do so by tracking adversarial perturbations in feature responses, allowing for automatic detection using average local spatial entropy. The method does not alter the original network architecture and is fully human-interpretable. Experiments confirm the validity of our approach for state-of-the-art attacks on large-scale models trained on ImageNet.
Optical music recognition using deep watershed detection
Lukas Tuggener, Ismail Elezi, Jurgen Schmidhuber, Thilo Stadelmann (ZHAW, University Venice, IDSIA)
In this talk, we introduce a novel object detection method, based on synthetic energy maps and the watershed transform, called Deep Watershed Detector (DWD). Our method is specifically tailored to deal with high resolution images that contain a large number of very small objects. Therefore it is well suited to process full pages of written music. We present state-of-the-art detection results of common music symbols and show DWD’s ability to work with synthetic scores equally well as with handwritten music.