EFFICIENT MEDICAL IMAGE ANALYSIS WITH LOW-BIT NEURAL NETWORKS

PhD Thesis Proposal Defence


Title: "EFFICIENT MEDICAL IMAGE ANALYSIS WITH LOW-BIT NEURAL NETWORKS"

by

Mr. Rongzhao ZHANG


Abstract:

In recent years, deep learning (DL) models have demonstrated stellar successes 
in various challenging medical image analysis applications. However, DL models 
are usually computationally intensive, which can impedes their applicability in 
intelligent healthcare products and platforms. In this thesis proposal, it is 
attempted to reduce the computation complexity of learning-based medical image 
analysis models by performing neural network quantization, which improves model 
efficiency by limiting the computational precision (i.e., bit-width) while 
maintaining the high accuracy.

First, a quantization-aware training (QAT) approach is designed for medical 
image segmentation tasks. This method features an adaptive quantization 
function with improved derivative approximation, the radical usage of residual 
connections as well as a knowledge distillation-aided training scheme, which, 
in combination, lead to highly accurate segmentation models with extremely low 
bit-width (e.g., binary or 2-bit models). Experiments on two public medical 
image segmentation datasets have demonstrated the efficacy of this method.

Second, to avoid the expensive finetuning phase in the training-based 
quantization x methods, a post-training quantization (PTQ) algorithm is 
developed, which requires neither a large-scale dataset, nor a long training 
stage. The method is based on a layer-wise optimization strategy and resorts to 
ADMM routines to solve each layer-wise problem efficiently. Extensive 
experiments have been carried out on both segmentation and registration tasks, 
and the results evidence its advantage over SOTA alternatives.

Third, the GPU implementation of low-precision CNN models is introduced, based 
on which an actual runtime analysis is presented on both individual 
convolutional layers and practical medical image segmentation models. A further 
study on the detailed time consumption of different model components is also 
performed, which verifies the theoretical complexity analysis in this thesis, 
and also reveals several future directions to improve the model acceleration 
performance.


Date:			Friday, 6 May 2022

Time:                  	4:00pm - 6:00pm

Zoom Meeting:
https://hkust.zoom.us/j/93789405390?pwd=aGJIV0w2NE4xend1VDFUdlFmOHg4dz09

Committee Members:	Prof. Albert Chung (Supervisor)
  			Prof. Chi-Keung Tang (Supervisor)
 			Dr. Dan Xu (Chairperson)
 			Prof. Weichuan YU (ECE)


**** ALL are Welcome ****