This study focuses on the critical need for bladder volume assessments in managing urinary disorders, highlighting ultrasound imaging (US) as a preferred noninvasive and cost-effective method. However, the reliance on professional expertise for interpreting US images presents a significant challenge due to its high operator dependency. 

To overcome this, the study introduces a deep learning-based system for bladder volume measurement tailored for point-of-care (POC) settings. This system utilizes a lightweight convolutional neural network (CNN)-based segmentation model, optimized for a low-resource system-on-chip (SoC). It efficiently detects and segments the bladder region in ultrasound images in real time, achieving high accuracy and robustness. Remarkably, it operates at 7.93 frames per second on the low-resource SoC, which is over 13 times faster than traditional networks, with minimal accuracy reduction. The practicality of this lightweight deep learning network was successfully tested using tissue-mimicking phantoms.

Summary:
This study developed a deep learning-based bladder volume measurement system for point-of-care use, employing a lightweight CNN model optimized for low-resource systems. This model significantly speeds up the processing of ultrasound images for bladder observation, maintaining high accuracy with minimal resource requirements. Its effectiveness was validated using tissue-mimicking phantoms, demonstrating its potential for widespread clinical application.

Rothband Comment:
As a company involved with imaging phantoms, we are particularly impressed with the advancements presented in this study. The development of a deep learning-based bladder volume measurement system optimised for point-of-care settings using a lightweight CNN model represents a significant leap in medical imaging technology. This innovation not only enhances the speed and accuracy of ultrasound imaging for bladder assessment but also does so with minimal resource requirements, which is crucial in diverse clinical environments. The successful validation of this system using tissue-mimicking phantoms, similar to our products, highlights the vital role of high-quality phantoms in the development and testing of cutting-edge medical imaging technologies. We see a great potential for collaboration, where our imaging phantoms could be used to further refine and test such innovative systems, ensuring they meet the highest standards of accuracy and reliability in various clinical scenarios.

https://www.kyotokagaku.com/

https://pubmed.ncbi.nlm.nih.gov/37237594/