Ultrasound (US) is a common non-invasive medical technique. Among all US techniques, the A-mode type receives less attention because the one-dimensional raw signal is difficult to interpret. However, besides B-mode’s non-invasive advantage, A-mode US is smaller, cheaper, more convenient, and easier to use, showing the potential to apply in daily portable & wearable scenarios, better to decipher the bone and muscle dynamics.
In this master thesis, the raw data will be revisited and explored using deep learning to unveil the unique features useful for medical applications. The exploration started from bone detection to muscle activity monitoring by interpreting raw US signals. The results showed high accuracy thanks to the universality, generalizability and robustness of the proposed deep-learning approaches. More attempts were made to define the weakness and scope of the technical performance, providing a clearer vision for the broader application in more domains of future developments.