Cerebro-vascular and heart diseases have together ranked 4th and 5th place in the 2022 top ten leading causes of death in Hong Kong, taking up more than 15% of the total in an unceasing trend. While conventional carotid ultrasound imaging is nothing short of comprehensive, it is highly operator-dependent and is worsened by the shortage of medical staff in Hong Kong.

The seemingly long queue for the expensive health screenings has put the high-risk groups, including but not limited to the elderly, in a vulnerable position as they can hardly perform regular and frequent check-ups.

In light of this, our team is determined to research a solution that is conducive to the preventive healthcare of strokes and cardiovascular diseases through one of the newly proposed devices: PyrocksTM Tag Lite.

This study aims to investigate an approach for developing a robust deep learning model for analysing ultrasound images and incorporate the model into our established prototype to perform intima-media thickness measurement and risk assessment.

Main points that the clinical trial can assist in solving the existing problem:

The acquisition procedures are non-invasive, painless, and safe for the participants. Clinical trials & test data will assist in testing and training our neural network model.

Inclusion Criteria:

• Adults (over the age of 18 years)(with Elderlies (over the age of 65 years) more preferred)
• Patients with cardiovascular diseases (CVD), including current smokers or diagnosed with diabetes, dyslipidaemia, coronary artery disease, cerebrovascular disease, hypertension, atherosclerotic cardiovascular disease, high blood pressure, high BMI index and those under antihypertensive treatment.

Exclusion Criteria:

• none

Based on Reference*, which utilized a sample of 50 subjects to assess the variation in IMT measurements among different observers and to reduce IMT variability through an automated computerized analyzing system, we have chosen a sample size of 80 subjects for our study on the automatic detection of carotid arteries and IMT measurement. This increased sample size is designed to account for potential exclusions due to unacceptable image quality, thereby ensuring that we maintain a sufficient number of high-quality images for analysis.

A total of 8000 images shall be obtained. Satisfactory images will be stored into the datasets for the deep learning model. By implementing the convolutional neural network through Tensorflow, the model will learn about the patterns and features of different image data: identify the carotid artery, measure the intima-media thickness and hence classify them into with or without cardiovascular risk according to risk charts.