This proposal is for an one-year project. In this project, we aim to investigate the feasibility of using AI for sonographic image interpretation. The main project is responsible for coordination between the two sub-projects and the main project, providing image resources, and using U-Net (Convolutional Networks for Biomedical Image Segmentation) and Transfer Learning to build up the models for image recognition and validating the efficacy of the models. The purpose of Subproject 1 is to develop an image recognition system for dynamic images: pericardial effusion. After building up the model, validating the efficacy and future revision will be done. Subproject 2 comes out an image recognition system for static images: hydronephrosis. After building up the model, validating the efficacy and future revision will be done.

Ultrasound is a non-invasive and non-radiated diagnostic tool in the emergency and critical care settings. In clinical practice, timely interpretation of sonographic images to facilitate decision-making is essential. However, it depends on operators' experience. As usual, it takes time for junior emergency physicians to have good diagnostic accuracy through traditional sonographic education. How to shorten the learning This proposal is for an one-year project. In this project, we aim to investigate the feasibility of using AI for sonographic image interpretation. The main project is responsible for coordination between the two sub-projects and the main project, providing image resources, and using U-Net (Convolutional Networks for Biomedical Image Segmentation) and Transfer Learning to build up the models for image recognition and validating the efficacy of the models. The purpose of Subproject 1 is to develop an image recognition system for dynamic images: pericardial effusion. After building up the model, validating the efficacy and future revision will be done. Subproject 2 comes out an image recognition system for static images: hydronephrosis. After building up the model, validating the efficacy and future revision will be done.

This pioneer study can provide two AI-assisted ultrasound image recognition systems in the real clinical conditions. They can experience of clinical applications and contribute to current medical education. Moreover, it can improve decision-making process and quality of care in the emergency and critical care units. Furthermore, the set-up models can be used in other target ultrasound image recognition in the future.

Inclusion Criteria:

• patients receiving echocardiography or renal ultrasound

Exclusion Criteria:

• patients not receiving echocardiography or renal ultrasound