Effective distinction of lipid microbubbles in pediatric bone phantom using a machine learning approach

This study investigates the ability to detect microbubble attenuation signals across a bone phantom (meant to mimic a pediatric femur) using ultrasound and machine learning. Lipid microbubbles were fabricated and injected through a hollow capillary fiber (200 μm) in a water tank attached to a syringe pump. Focused ultrasound (1.1 MHz, 1–10 MPa, 500 pulses) was applied to the flowing bubbles in the presence or absence of a custom 3-D-printed bone phantom. A co-aligned passive cavitation detector (PCD) captured the reflected ultrasound signals. Ultrasound signals were compared using Fast Fourier Transform (FFT) to process the data. Eleven spectral features were extracted and used as the basis for machine learning algorithms. The results confirm that despite strong signal interference from the bone, machine learning can effectively distinguish the presence of lipid microbubbles. The study signifies the potential of lipid microbubbles for targeted drug delivery to bone marrow using PCD, a method often deemed challenging due to the hard barrier posed by bone structures. This method has important implications for medical treatments such as chemotherapy or gene therapy, which often struggle to reach the bone marrow due to its protective nature.