The full name of MICCAI is International Conference on Medical Image Computing and Computer Assisted Intervention; it is a top international conference in the field of medical image analysis. The 23rd MICCAI conference will be held in Lima, Peru in October 2020. The conference received more than 2,900 submissions, and the acceptance rate in previous years was usually less than 30%.
1. Q. Yao, Z. He, H. Han, and S. Kevin Zhou, Miss the point: Targeted adversarial attack on multiple landmark detection.
Recent methods in multiple landmark detection based on deep convolutional neural networks (CNNs) reach high accuracy and improve traditional clinical workflow. However, the vulnerability of CNNs to adversarial-example attacks can be easily exploited to break classification and segmentation tasks. This paper is the first to study how fragile a CNN-based model on multiple landmark detection to adversarial perturbations. Specifically, we propose a novel Adaptive Targeted Iterative FGSM (ATI-FGSM) attack against the state-of-the-art models in multiple landmark detection. The attacker can use ATI-FGSM to precisely control the model predictions of arbitrarily selected landmarks, while keeping other stationary landmarks still, by adding imperceptible perturbations to the original image. A comprehensive evaluation on a public dataset for cephalometric landmark detection demonstrates that the adversarial examples generated by ATI-FGSM break the CNN-based network more effectively and efficiently, compared with the original Iterative FGSM attack. Our work reveals serious threats to patients’ health. Furthermore, we discuss the limitations of our method and provide potential defense directions, by investigating the coupling effect of nearby landmarks, i.e., a major source of errors in our experiments.
2. H. Li, H. Han, and S. Kevin Zhou, Bounding maps for universal lesion detection.
Universal Lesion Detection (ULD) in computed tomographyplays an essential role in computer-aided diagnosis systems. Many detec-tion approaches achieve excellent results for ULD using possible bound-ing boxes (or anchors) as proposals. However, empirical evidence showsthat using anchor-based proposals leads to a high false-positive (FP)rate. In this paper, we propose a box-to-map method to represent abounding box with three soft continuous maps with bounds in x-, y-and xy- directions. The bounding maps (BMs) are used in two-stageanchor-based ULD frameworks to reduce the FP rate. In the 1 st stage ofthe region proposal network, we replace the sharp binary ground-truthlabel of anchors with the corresponding xy-direction BM hence the pos-itive anchors are now graded. In the 2 nd stage, we add a branch thattakes our continuous BMs in x- and y- directions for extra supervision ofdetailed locations. Our method, when embedded into four state-of-the-art two-stage anchor-based detection methods, brings a free detectionaccuracy improvement (e.g., a 1.68% to 3.85% boost of sensitivity at 4FPs) without extra inference time.
3. Z. Huang, Y. Ding, G. Song, L. Wang, R. Geng, H. He, S. Du, X. Liu, Y. Tian, Y. Liang, S. Kevin Zhou, and J. Chen, BCData: A large-scale dataset and benchmark for cell detection and counting.
4. Y. Lyu, W. Lin, H. Liao, J. Lu, and S. Kevin Zhou, Encoding metal mask projection for metal artifact reduction in computed tomography.
5. W. Wang, Q. Song, J. Zhou, R. Feng, T. Chen, W. Ge, D.Z. Chen, S. Kevin Zhou, W. Wang, and J. Wu, Dual-level selective transfer learning for intrahepatic cholangiocarcinoma segmentation in non-enhanced abdominal CT.