Adaptive Video Encoding And Streaming Of Ultrasound Videos

Abstract

Medical Video Applications have become an integral component of medical healthcare, particularly in areas such as remote diagnoses, telemedicine and medical education. Videos from various modalities, including the ultrasound modality, are essential for detecting and evaluating critical medical conditions. However, securing the quality of the communicated video in real time presents serious difficulties due to the dynamic, time-varying nature of the wireless channels. The video systems must swiftly adjust to varying bandwidths while ensuring the quality of the communicated video. To address these challenges, in this study, we have developed Forward Prediction Models for video quality, encoding frames per second (fps) and bitrate demands, alongside implementing a multi-objective optimization framework for real-time video encoding adaptation. The method satisfies the time-varying constraints and is validated using two different encoders (x265 and SVT-AV1). The aim is to enhance video quality while reducing encoding time and the required bitrate. Forward Prediction Models are built via offline training on many distinct video compression instances, per optimization goal (bitrate, video quality, encoding fps). The methodology uses actual network traces conducted over 3G wireless networks. An adaptive controller is then implemented to adapt to instantaneous bandwidth fluctuations and initiate encoding adaptations using the generated Forward Prediction Models. The controller triggers an encoding configuration switch to match the time-varying wireless network state. For validation, a dataset of CCA ultrasound videos is used, with a resolution of 560x448 at 40 fps. This study evaluates Forward Prediction Models for adaptive video encoding, focusing on robustness, error distribution, and comparative performance. The median percentage fluctuations for coefficients and the adjusted R2 of the fitted models remained below 5% and 0.7, respectively, indicating model resilience. A comparison with the traditional HTTP Adaptive Streaming (HAS) algorithm revealed that Forward Models offered better video quality (VMAF and SSIM) and reduced buffering incidents and video stalling, especially as the InTransit value increased. Buffer utilization statistics highlighted the effectiveness of the Forward Model in maintaining buffer fullness, minimizing rebuffering and enhancing Quality of Service (QoS) metrics. Overall, Forward Prediction Models proved effective for real-time adaptive video streaming applications, including medical applications and ultrasound contexts, offering notable improvements in video quality, bitrate demands and user Quality of Experience (QoE). Finally, an adaptive video streaming system based on MPEG-DASH was implemented, where medical and especially CCA videos play in good quality without buffering effects. This system improves medical education, telemedicine, doctor-to-doctor communication and remote diagnosis.