[阅读笔记]Diagnostic Image Quality Assessment and Classification in Medical Imaging: Opportunities and C

Abstract:

这个任务是判断MRI的质量好不好 accessing medical image quality and detecting diagnostic and non-diagnostic images

Introduction

Coupled with physiological and patient motion, MR images often suffer from many forms of motion artifacts, resulting in non-diagnostic quality images 

1.技师要审查有没有质量问题。2.患者后来又会被要求重拍。

以上两点就是这个工作的multivation。

 detection and classification tasks in medical imaging are unique and particularly challenging for several reasons, which include: i) the lack of clear discriminant between a diagnostic and non-diagnostic quality image, resulting in subjectivity in classification, ii) unbalanced data sets, and iii) the lack of sufficient training data and reliable labels

in this work, we closely investigate the challenges in the detection and classification task for medical image diagnostic quality assessment. We present our results on two DL models, compare their performance, and investigate several characteristics of the data set.

Methods

2.2. Model Architecture

比较了两个结构：C1) A simple 4-layer convolutional neural network and C2) A standard convolution kernel-based ResNet-10 architecture。我们的分析认为非线性性在这个任务里是不重要的。

Experiments and Results

3.1. Model Architecture Analysis

用这个软件"Reproduction of motion artifacts for performance analysis of prospective motion correction in MRI"生成有问题的图。

低层特征差异明显，高层差异不明显。所以我们要用更简单的网络。

 

后面没看。