基于卷积神经网络的胶囊内镜息肉与溃疡辅助诊断

针对目前胶囊内镜(WCE)自动检测方法需要对每种病灶设计对应的识别算法以及识别准确率不高的问题,设计一种基于卷积神经网络的息肉与溃疡辅助诊断算法。与传统检测算法相比,卷积神经网络可自动学习病灶图像特征,实现更强泛化能力,更高准确率和效率。该方法针对具体WCE图像,首先评价图像R、G、B通道携带信息的特征;其次,分析全局直方图均衡化、伽玛变换和拉普拉斯变换对提升图像对比度的效果,选择其中表现最佳者与信息最丰富的2个颜色通道组合成3通道输入到卷积网络中训练和识别。测试表明,本算法识别准确率96.8%,比传统的经典图像检测方法高出至少16.73%,检测速度达到68.6图/s,能够推广应用到医疗辅助诊断领域。

Auxiliary diagnosis of capsule endoscopy on polyps and ulcers based on convolutional network

In view of the problems of current capsule endoscopy (WCE) automatic detection method needs to design the corresponding identification algorithm to each lesions design and the recognition accuracy is not high, a polyps and ulcer-assisted diagnosis method based on convolutional neural network is proposed. Compared with the traditional detection algorithm, the convolutional neural network can automatically extract different lesions characteristics, with stronger robustness, higher accuracy and efficiency. This method is aimed at the specific WCE image to first compare the characteristics of the image R, G, B channel carrying information. Secondly, the effect of global histogram equalization, gamma transformation and Laplace transformation to improve image contrast is analyzed, and the two color channels with the best performance and the most informative are combined into 3 channel inputs. Enter training and recognition into the convolution network. The test results tell that the average accuracy rate of this algorithm is 96.80%, which is at least 16.73% higher than the traditional classical image detection method, and the detection speed reaches 11 graph/s, which can be applied to the field of medical auxiliary diagnosis.