e-health and resource management scheme for a deep learning-based detection of tumor in wireless capsule endoscopy videos

Recently, a lot of concentration is on how early diagnosis for critical diseases can be accommodated with deep learning (DL). e-health is an emerging area in the junction of medical informatics, public health, and business, indicating health assistance and data delivered or improved by the Internet and associated technologies. Resource management as bandwidth allocation problem is a key problem while transmitting processed medical data where both data integrity and quality are of utmost importance. To address the early intelligent detection and diagnosis of the diseases, an end-to-end DL model i.e., You Only Look Once (YOLOv3-tiny) is selected for the detection of the tumor within with wireless capsule endoscopy videos. The DL mode is an improved version of the YOLOv3-tiny wherein each convolutional layer, different convolutional filters, is employed to extract both local and global features. The motivation is early detection of the critical disease followed by remote physician diagnosis where resource management as a bandwidth allocation is investigated using encoders like H.265/HEVC and VP9. The proposed scheme controls the frame rate, video resolution, and compression ratio as quantization based on the intelligent decision from the DL model. The performance of the improved YOLOv3-tiny model is benchmarked with YOLOv3-tiny and our previous work in terms of precision, sensitivity, F1-score, and F2-score. Furthermore, the resource management results are shown in terms of bandwidth and storage for both encoders.