Tele-sonography addresses the shortage of sonographers by allowing the semi-skilled persons to scan the patients and transmit the scanned data to the cloud for analysis. In general, the ultrasound scanners generate large volumes of data due to the subjectivity of the scanning and the semi-skilled nature of the person. Transmitting all the non-representative data to the cloud will result in network congestion, and also makes the diagnostics inefficient and unreliable. Considering the circumstances, there is a need for an automated algorithm to assist the semi-skilled persons to filter the non-representative data getting into the cloud for analysis, as well as to scan the representative data for diagnosis. To address this issue, we propose a Convolution Neural Network (CNN) based algorithm for classifying representative and non-representative data for diagnosis. For analysis, we considered the ultrasound images corresponding to the liver. The proposed algorithm classified representative and non-representative images of the liver with an accuracy of 99.21%. The proposed algorithm can be used to filter the non-representative getting into the cloud for analysis as well as in the development of Computer-Aided Diagnostic (CAD) algorithms.

中文翻译：

---

基于CNN的框架，可用于CAD和远程超声检查应用的肝脏图像代表性检测

远程超声检查通过允许半熟练人员扫描患者并将扫描的数据传输到云中进行分析，从而解决了超声检查人员的不足。通常，由于扫描的主观性和人的半熟练性质，超声扫描仪会生成大量数据。将所有非代表性数据传输到云将导致网络拥塞，并使诊断效率低下且不可靠。考虑到这种情况，需要一种自动算法来协助半熟练人员过滤进入云中的非代表性数据进行分析，以及扫描代表性数据进行诊断。为了解决这个问题，我们提出了一种基于卷积神经网络（CNN）的算法，用于对代表性和非代表性数据进行分类以进行诊断。为了进行分析，我们考虑了与肝脏相对应的超声图像。所提出的算法对肝脏的代表图像和非代表图像进行了分类，准确性为99.21％。所提出的算法可用于过滤非代表进入云进行分析以及在计算机辅助诊断（CAD）算法的开发中使用。