Neural network-enhanced color photoacoustic imaging using multi-frequency PMUT array

This study investigates colored photoacoustic imaging (PAI) scan by integrating a neural-network image classification algorithm with a multi-frequency Piezoelectric Micromachined Ultrasound Transducer (PMUT) array. Fabricated on an AlN-on-SOI platform, the PMUT array features 133 (19 × 7), 196 (28 × 7), and 246 (41 × 6) transducers of distinctive designs, targeting under-liquid resonant frequencies of 760 kHz, 1.17 MHz, and 1.65 MHz respectively. This multi-frequency capability broadens the range of detectable photoacoustic signals, enhancing sensitivity to variations in acoustic responses associated with the heat absorption properties of different colored targets. The neural network, initially trained on extensive datasets from stationary colored pencil leads, achieved over 99 % accuracy in color classification. When integrated with a 2D scanning and image reconstruction system, this setup enabled comprehensive color PAI scans of phantoms embedded with colored pencil leads in random sequences. These advancements extend PAI’s diagnostic capabilities beyond that of traditional ultrasound transducers, offering enhanced resolution and further insights into structural materials including biomedical applications. • Multi-frequency PMUT array uses custom transducers for enhanced PAI resolution. • Broadband PMUT array detects and amplifies subtle absorption variations. • On-chip processing extracts spectral patterns for neural-network classification. • Neural network classification exceeds 99 % accuracy in color 2D PAI scans. • Method identifies color-based absorption differences for advanced biomedical insights.