Emma Wikberg: Optimizing imaging for improved cancer treatment
Radioactive substances play a crucial role in effective cancer treatment. They are not only used to target and destroy tumors but also to detect cancer using a SPECT camera in combination with a CT scanner. This thesis demonstrates that, with the help of AI, it is possible to shorten imaging time while maintaining both image quality and the ability to calculate radiation doses from the images, referred to as dosimetry.
EMMA WIKBERG
Dissertation defense: 11 April 2025 (click for details)
Doctoral Thesis: Evaluations of Monte Carlo and AI Techniques for SPECT Reconstruction
Research area: Medical Radiation Sciences
Sahlgrenska Academy, The Institute of Clinical Sciences
Early detection of cancer can significantly impact patient survival. By injecting a radioactive substance linked to a tracer into the body, certain types of tumors can be identified. In higher doses, radioactive substances can also be used to irradiate tumors from within, providing an internal form of radiation therapy.
This thesis explores SPECT imaging both for diagnosis and post-therapy assessment. During diagnostics, the goal is to minimize radiation exposure to patients to reduce the risk of stochastic effects such as cancer while still collecting enough data, limiting the noise, to ensure sufficient image quality for tumor detection. Conversely, in therapeutic applications, the aim is to maximize radiation dosage for effective tumor treatment while avoiding damage to healthy organs. However, treatments using radioactive drugs are rarely curative on their own.
"Currently, these treatments are generally administered as standardized procedures with fixed doses and cycles. One of the major objectives in this field is to personalize treatments for better outcomes. Achieving this requires a deeper understanding of radiation doses delivered to both tumors and at-risk organs—knowledge that can be gained through improved dosimetry," explains Emma Wikberg, medical physicist at the Nuclear Medicine Radiation Physics Unit at Sahlgrenska University Hospital and a doctoral student at the Institute of Clinical Sciences.
Enhancing SPECT imaging quality
When using radioactive substances for diagnostics or treatment, they are attached to a targeting molecule that binds to tumor cells to a higher degree than to normal tissue. Imaging is performed using a SPECT (Single Photon Emission Computed Tomography) camera, combined with a CT scanner, to visualize tumors for diagnosis, treatment planning, or treatment evaluation.
"The image quality of SPECT cameras is inherently limited due to the physical properties of radiation and the design of the detectors. In diagnostics, we cannot administer unlimited amounts of radioactivity, as we must account for the risks of ionizing radiation. Additionally, patients cannot remain still indefinitely, restricting the amount of data we can collect. These factors contribute to noisy images in both diagnostic and post-therapy scans."
The research presented in this dissertation focuses on improving SPECT image quality using advanced reconstruction methods. By simulating various physical effects of radiation and detector response, it is possible to correct for these limitations using Monte Carlo simulations.
"Furthermore, we have implemented AI to reduce imaging time, allowing us to capture more images and cover a larger portion of the patient’s body. We have utilized an artificial neural network, specifically a convolutional neural network (CNN), which is particularly well-suited for medical imaging applications. Typically, SPECT cameras capture images from 120 different angles, but our network can infer the remaining 90 angles using only every fourth collected projection."
Refinements that can improve quality of life and survival rates
"We have demonstrated that Monte Carlo-based reconstruction provides superior image quality compared to standard reconstruction techniques. Additionally, we have shown that AI can reduce imaging time by a factor of four while maintaining image quality and reliable dosimetry (calculations of radiation doses)."
This advancement enables the collection of more images, covering a greater area of the patient’s body, thereby allowing for dose calculations for all at-risk organs.
"By performing dosimetry, we can better map and evaluate treatments. In the future, this will help us personalize therapies, which is expected to enhance both patient quality of life and survival rates."
Text: Susanne Lj Westergren
The illustration in its entirety was composed by Emma Wikberg, with the following representative elements from the thesis:
Upper left – the CUSIP network architecture.
Center – dual-head SPECT camera (personal photo using AI filter from fotor.com).
Right – circles representing projections where every fourth projection is original (acquired) and remaining 90 are synthetic intermediate projections (generated with the CUSIP network).
Background – photons simulated with Monte Carlo on a photograph of ice. Monte Carlo simulations by Jens Hemmingsson.