面向眼底图像小目标检测的无监督学习方法

小目标检测是图像处理领域的一个难点,尤其是医学图像中的小目标检测。微动脉瘤MA作为眼底图像中的一类小目标,尺寸小、局部对比度较低,并且存在较多的噪声干扰,检测难度较大。传统的检测方法需要手工提取特征,难以准确检测MA。而基于深度学习的检测需要进行复杂的前期准备工作,工作量大,并且难以解决正负样本数量不平衡的问题,容易产生过拟合。稀疏编码器SAE是一种无监督机器学习算法,可以在样本数量不平衡的环境中有效地提取样本的特征。因此,提出了一种基于SAE的无监督学习方法检测MA,采用反向传播更新SAE的权重和偏置以提取样本的特征,并利用提取的特征训练Softmax,最终实现MA的准确检测。为验证方法性能,选取了Retinopathy Online Challenge、DIARETDB1和E-ophtha-MA 3个数据库分别进行实验。实验结果表明,本文方法能够准确地检测出眼底图像中的MA,并且获得了较高的准确率和灵敏度。准确率分别为98.5%,87.2%和92.6%,灵敏度分别为99.9%,99.8%和98.7%。

Unsupervised learning for smallobjects detection in retinal images

Small object detection is an unresolved problem for image processing, especially for medical image processing. Microaneurysm (MA) is a kind of small objects in retinal images. It has small size, low local contrast, and more noise interference, so it is difficult to be detected.Traditional detection methods require manual extraction of features, making it difficult to accurately detect MA.The detection based on deep learning requires a large amount of complex preparatory work, and it is difficult to solve the imbalance problem between positive and negative samples, which is easy to cause over-fitting.Sparse autoencoder (SAE) is an unsupervised machine learning algorithm that efficiently extracts the features of samples in an environment with unbalanced sample data.Therefore, an unsupervised learning method based on SAE is proposed to detect MA. The weights and offsets of SAE are updated by backpropagation to extract the features of the samples, and the extracted features are used to train softmax to achieve accurate detection of MA.In order to evaluate the performance of the method, three databases (Retinopathy Online Challenge, DIARETDB1 and E-ophtha-MA) are used to carry out experiments.Experimental results show that the method can accurately detect MA in retinal image and obtain higher accuracy and sensitivity. The accuracy rates are 98.5%, 87.2%, and 92.6% respectively, and the sensitivity are 99.9%, 99.8%, and 98.7% respectively.