Design of optoacoustic imaging system for deep penetration medical diagnostics

Abstract

This final year project is dedicated to the objective of improving photoacoustic imaging system using Artificial Intelligence algorithms capable of reconstructing images. Theoretical underpinnings of the selected algorithms, their practical implementation, and ensuing results were thoroughly discussed. To begin, the project underscored the importance of a non-invasive technique for early stage cancer diagnostics, exemplified by Photoacoustic Imaging. The operational principle of a Photoacoustic Microscopy System was elucidated, along with the process of signal conversion into images. Furthermore, the project explored the use of convolutional neural networks such as U-Net, Multi-Res U-net, U-net with Attention and Multi-Res U-net with attention to enhance image reconstruction qualities. The architecture of these networks and their respective advantages were elucidated. Subsequently, the networks were constructed, employing an input dataset comprising 1000 pre-processed photoacoustic images. Model performance was evaluated against a test dataset using metrics such as Peak Signal-to-Noise Ratio and Structural Similarity Index Measure. Additionally, the reconstructed images generated by the models were compared. Based on the obtained results, the optimal reconstruction algorithm was proposed.