| Proj No. | A2154-251 |
| Title | Generative artificial intelligence for sub-surface diffuse optical imaging |
| Summary | Wound characterization and monitoring are critical components of clinical care, yet existing methods often fail to deliver timely, accurate, and non-invasive assessments. Traditional approaches, such as visual inspection and biopsies, can be subjective and frequently lack the depth resolution to detect subtle sub-dermal changes. Moreover, chemical-based methods, although sometimes used to enhance diagnostic capabilities, can have adverse side effects, limiting their utility and safety in vulnerable patient populations. This underscores the pressing need for innovative imaging techniques that are both non-invasive and highly sensitive. Diffuse optical imaging offers a compelling alternative by employing near-infrared light to probe tissue properties beneath the surface, providing insights into the physiological state of tissues with the benefits of being non-invasive and cost-effective. Its ability to reveal sub-surface tissue details makes it particularly valuable for assessing wound healing and detecting infections or lesions that may elude standard diagnostic tools. This project proposes an innovative framework integrating diffuse optical imaging with generative artificial intelligence to perform sub-surface imaging of human skin and infant heads. In this approach, conditional variational autoencoders are employed to reconstruct the optical property distributions of tissues, enabling precise wound classification in the skin and early detection of sub-dermal head injuries and lesions in infants. By reconstructing these optical properties, clinicians will be equipped to monitor wound healing and identify infections that are not visible to the naked eye. This method broadens the diagnostic capabilities beyond traditional imaging techniques and offers a non-invasive alternative for patients who cannot undergo MRI. The successful candidate will generate digital phantoms that closely mimic the optical characteristics of human tissue and simulate light propagation through them using Monte Carlo methods. Building on these simulations, the candidate will simulate a diffuse optical imaging setup to compile a comprehensive dataset of optical measurements. This dataset will be used to train a conditional variational autoencoder that will both reconstruct the spatial distribution of optical properties and classify tissue pathology based on the acquired data. This project represents an interesting opportunity at the intersection of biomedical imaging and artificial intelligence, promising to enhance diagnostic accuracy and improve patient outcomes in challenging clinical scenarios. |
| Supervisor | Ast/P Matthew Foreman (Loc:S1 > S1 B1C > S1 B1C 77, Ext: ?) |
| Co-Supervisor | - |
| RI Co-Supervisor | - |
| Lab | Photonics I (Loc: S1-B3a-08) |
| Single/Group: | Single |
| Area: | Microelectronics and Biomedical Electronics |
| ISP/RI/SMP/SCP?: |