Noise reduction using deep learning for quantitative SPECT reconstruction to personalise molecular radiotherapy treatments
LEAD SUPERVISOR: Prof Geoff Higgins, Department of Oncology
Co-supervisor: Dr Daniel McGowan, Department of Oncology
Commercial partner: Hermes Medical Solutions Ltd, London
The quality of medical images used for cancer diagnosis and treatment is degraded during the image reconstruction process, reducing diagnostic accuracy. In Nuclear Medicine, tumours can be imaged by injecting oncology patients with a small amount of a radioactive pharmaceutical taken up by tumours. The distribution of radiopharmaceutical inside the patient is imaged by a single photon emission computed tomography (SPECT) scanner and the resulting images are used to report the tumour size and other characteristics, to inform patient management.
Targeted radiopharmaceuticals or ‘theranostics’ can be given for therapeutic rather than imaging purposes. In Oxford, Lu-177 DOTA-peptides are administered to patients with neuro-endocrine tumours and subsequently imaged with quantitative SPECT to calculate the radiation dose delivered to the tumours. This step is crucial for optimising and personalising the treatment to the patient, giving the best chance of a successful outcome.
Recently it has been made a legal requirement to plan and verify the radiation dose delivered to tumours and potential organs at risk. This is a high impact area since novel radiopharmaceuticals are expected to be used much more widely (for example Lu-177 PSMA has recently been shown to be significantly more effective for prostate cancer compared to chemotherapy).
SPECT images are reconstructed from the acquired data using specialist algorithms. Clinically, the reconstruction method used is degraded by noise and using current methods to remove the noise (denoising) leads to image blurring and so affect quantitation. Deep learning (DL) offers many possibilities for novel approaches to denoising, including DL priors as well as post-reconstruction methods. The project would explore the potential of DL to improve SPECT image quality, facilitating more accurate imaging for cancer diagnosis. The possibility for significantly reducing scan time would also be explored, improving patient experience and comfort, as well as enabling the legally mandated dosimetry analysis.
Apply using course: DPhil in Oncology