Hardware Aware Learning for
Medical Imaging and Computer Assisted Intervention

8:00am–12:00pm, Oct. 17, 2019, Shenzhen, China

Technical Program

8:00 - 8:10 Opening remarks
8:10 - 8:50 Keynote: Deep Learning and Medical Image Applications: Current Challenges and New Approaches
Danny Chen
University of Notre Dame
8:50 - 9:00 Break
9:00 - 9:25 ANT-UNet: Accurate and Noise-Tolerant Segmentation for Pathology Image Processing
Yufei Chen1, Qinming Zhang1, Tingtao Li2, Hao Yu2, Mei Tian1, Cheng Zhuo1
1Zhejiang University, 2Southern University of Science and Technology
9:25 - 9:50 Fixed-Point U-Net Quantization for Medical Image Segmentation
MohammadHossein AskariHemmat1, Yvon Savaria1, Jean-Pierre David1, Sina Honari2,
Christian Perone1, Lucas Rouhier1, Julien Cohen-Adad1
1Polytechnique Montreal, 2MILA
9:50 - 10:15 An Analytical Method of Automatic Alignment for Electron Tomography
Shuang Wen, Guojie Luo
Peking University
10:15 - 10:30 Break
10:30 - 10:55 Deep Compressed Pneumonia Detection for Low-Power Embedded Devices
Hongjia Li1, Sheng Lin1, Ning Liu1, Caiwen Ding2, Yanzhi Wang1
1Northeastern University, 2University of Connecticut
10:55 - 11:20 Hardware Acceleration of Persistent Homology Computation
Fan Wang1, Chunhua Deng2, Bo Yuan2, Chao Chen1
1Stony Brook University, 2Rutgers University
11:20 - 11:55 D3MC: A Reinforcement Learning based Data-driven Dyna Model Compression
Jiahui Guan, Ravi Soni, Dibyajyoti Pati, Avinash Gopal, Venkata R Saripalli
GE Healthcare

Organizing Team

Program Committee Chair: Yiyu Shi University of Notre Dame
Program Committee Co-Chairs: X. Sharon Hu University of Notre Dame
Danny Chen University of Notre Dame
Publicity Chair: Bei Yu Chinese University of Hong Kong

Program Committee Members:

Albert Liu Kneron
An Zeng South China University of Technology
Guojie Luo Peking University
Jian Zhuang Guangdong General Hospital
Jianxu Chen Allen Institute
Jinjun Xiong IBM T. J. Watson Research Center
Lichuan Ping Max BioE
Steve Jiang University of Texas Southwestern Medical Center
Umamaheswara Rao Tida North Dakota State University
Yongpan Liu Tsinghua University
Yu Wang Tsinghua University

Workshop Description

With the prevalence of deep neural networks, machine intelligence has recently demonstrated performance comparable with, and in some cases superior to, that of human experts in medical imaging and computer assisted intervention. Such accomplishments can be largely credited to the ever-increasing computing power, as well as a growing abundance of medical data. As larger clusters of faster computing nodes become available at lower cost and in smaller form factors, more data can be used to train deeper neural networks with more layers and neurons, which usually translate to higher performance and at the same time higher computational complexity. For example, the widely used 3D U-Net for medical image segmentation has more than 16 million parameters and needs about \(4.7 \times 1013\) floating point operations to process a \(512 \times 512 \times 200\) 3D image. The large sizes and high computation complexity of neural networks have brought about two emerging issues that need to be addressed by the joint efforts between hardware designers and researchers in the MICCAI society towards hardware aware learning.

First, when powerful computing resources are easily accessible through network connection, such large networks may not impose significant challenges. However, for many medical applications where latency, privacy or reliability is critical (such as implantable medical devices or health monitoring), inference has to be done locally, and such computation is subject to stringent area and power constraints due to limited resources available. To address the computational demands, hardware designers have started to explore techniques to compress deep neural networks for efficient local inference (i.e., edge inference). The ultimate judgment of such techniques is that lower power and area overhead can be achieved with minimal loss in inference accuracy. As many researchers in the MICCAI society are focusing on increasing inference accuracy through designing more complex networks, a correlated race exists between these researchers and hardware designers. Semiconductor technology scaling based on Moore's law has provided hardware designers a relatively easy path towards accommodating increasing network sizes. However, with the slowdown of the scaling trends, a clear gap between hardware capacity and computational demand has emerged. It would therefore be of interest to explore various hardware designs coupled with algorithm innovations that could help to bridge the gaps.

Second, in many real-time medical applications such as image/VR guided surgery, it is desirable to further accelerate the inference speed of large neural networks. As hardware acceleration of deep neural networks is popular today, it would be interesting to explore customized hardware that utilizes the dedicated structure of a network for maximal efficiency. On the other hand, for a given hardware platform, it would also be interesting to see how to design a neural network/algorithm that can work best with it for a particular medical application. Ultimately, the joint exploration/co-design of hardware and neural networks can benefit many important problems within the MICCAI scope.

This workshop intends to build a bridge between hardware designers and researchers in the MICCAI society and provides a forum for idea exchanges that can lead to interdisciplinary collaborations.

Invited Speakers:

To be provided soon

Important Dates

Submission due: 06/15/2019 06/30/2019
Notification: 07/04/2019
Camera-ready paper due: 07/25/2019

Submission link:

Manuscript Format

Papers should be submitted electronically following the guidelines for authors and LaTeX and MS Word templates available at Lecture Notes in Computer Science, double blind review). Manuscripts should be up to 8-pages. No modifications to the templates are permitted. Failure to abide by the formatting guidelines will result in immediate rejection of the paper.


Accepted workshop papers will be part of the MICCAI Satellite events joint Springer Lecture Notes in Computer Science (LNCS) proceedings. The authors will also have a chance to extend their work by at least 30% and submit to a special issue of a journal such as IEEE Transactions of Computer-Aided Design of Integrated Circuits and Systems (TCAD).