Page 32 - ARNM-2-2

Advances in Radiotherapy
& Nuclear Medicine Generative AI enhances radiotherapy

fostering innovative and efficient solutions for a more specific tumor and beam types, enhancing the diversity of
environmentally friendly and sustainable future. training data for dose prediction tasks. Additional studies
33
have aimed to improve dose distribution across different
3. The convergence of GAI and generative organs while minimizing radiation exposure to healthy
CS with CR tissues, achieving state-of-the-art performance in terms of
target coverage and organ sparing. 34
3.1. The use of GANs in radiotherapy
In radiotherapy planning, creating an ideal treatment
In the ever-evolving landscape of medical technology, the plan involves a collaborative effort between dosimetrists
integration of AI has paved the way for groundbreaking and oncologists, requiring iterative adjustments to beam
advancements in treatment planning for radiotherapy. parameters. This process is both time-consuming and
35
Among these innovations, GANs stand out as a labor-intensive, and the quality of the plan often depends
transformative tool, offering novel approaches to optimize on the expertise of the dosimetrists, leading to variability in
treatment planning processes. GANs, with their ability to standards and potentially sub-optimal plans. To address
35
generate synthetic data and harness complex patterns, are these challenges, researchers are exploring methods to
revolutionizing how treatment plans are formulated and automatically predict the outcome of radiotherapy plans,
executed. specifically the distribution of radiation dose across
GANs offer a revolutionary approach to optimizing the patient’s anatomy. By predicting the dose map,
36
treatment planning in radiotherapy by addressing the dosimetrists can quickly assess the plan’s effectiveness and
challenges of multimodal medical imaging. While make necessary adjustments, reducing the trial-and-error
26
obtaining multiple imaging modalities for diagnosis and process and overall planning time. 36
planning presents practical limitations, GANs enable cross- One common approach for automatic dose prediction
mode image synthesis, extracting crucial information from is known as knowledge-based planning (KBP), which
a single modality. 27 utilizes past patient data with similar anatomical
37
Particularly in radiotherapy planning, where positron characteristics to inform treatment plans for new patients.
emission tomography is essential but magnetic resonance However, traditional KBP methods focus on predicting
imaging (MRI) offers superior delineation, GANs facilitate dosimetric endpoints or dose volume histograms based on
MRI treatment planning by generating synthetic computed handcrafted features, which may not fully capture spatial
tomography (CT) images from MRI data. 27,28,38 information. 37,38 To overcome this limitation, researchers
are turning to GANs to approach dose prediction as an
Various GAN-based methods, such as the integration image synthesis task. GANs have shown promise in
39
of dense blocks and the use of mutual information loss generating realistic dose distributions from input data,
functions, have demonstrated remarkable accuracy in such as CT scans and masks of target and surrounding
generating CT images, minimizing systematic errors, and organs. 40
facilitating MRI workflows for treatment planning. 27,30
Furthermore, GANs enhance volume segmentation and Moreover, efforts have been made to accelerate dose
dose performance, making them invaluable tools for calculation processes using GANs. By leveraging innovative
optimizing radiotherapy planning. 31 architectures and loss functions, researchers have enabled
the real-time generation of fluence maps from CT scans,
Studies also highlight the feasibility of generating reducing the time required for treatment planning without
CT images from multi-sequence MRI data, showcasing compromising accuracy. In addition, GANs have been
41
GANs’ adaptability and efficiency in diverse clinical employed to generate dose distributions with comparable
scenarios. 31,32 The rapid generation of CT images by GANs accuracy to traditional Monte Carlo simulations but
underscores their potential for integration into MRI- with significantly reduced computational time. These
41
guided radiotherapy workflows, offering both accuracy advancements hold promise for more efficient and accurate
and speed. 32 treatment planning processes, offering practical benefits
The field of radiotherapy planning has undergone a for both clinicians and patients.
significant transformation with the introduction of deep In the realm of relative stopping power prediction,
learning techniques, particularly GANs. Researchers GANs have facilitated the prediction of critical parameters
have developed various GAN-based methods to address from cone beam CT scans, enabling the implementation
challenges, such as limited dataset availability and time- of adaptive treatment planning strategies with enhanced
consuming dose calculations. For instance, some studies precision and efficiency. Adaptive radiotherapy enables
32
have focused on synthesizing dose distributions tailored to adjustments to the treatment plan to optimize the delivery

Volume 2 Issue 2 (2024) 4 doi: 10.36922/arnm.3523