New artificial intelligence (AI) capabilities may make it possible to improve the effectiveness of brachytherapy for men with prostate cancer (PCa) by almost instantly generating dosage plans, according to investigators.
In a typical high-dose rate (HDR) brachytherapy procedure for PCa, needle applicators are first inserted by the physician to the tumor target. A planner then develops a treatment plan manually. During this time the patient carries the needles, waiting for the planning to finish. With the current standard of care, it takes up to an hour or more to generate a high-quality plan. New research, however, suggests that this wait time can be substantially reduced with enhanced deep-learning models that streamline the process down to about 10 minutes.
Currently, a physician has to manually outline the
tumor target that is to be treated as well as the organs nearby. A planner has
to manually digitize needles and iteratively tune treatment planning
hyperparameters of the plan optimization to arrive at a planner acceptable
dose. Next, feedback loops between the physician and planner are required to
determine the acceptable dose.
“Achieving optimal treatment plans in near real-time
is important and part of a new broader mission to use AI is to improve all
aspects of cancer care,” said Xun Jia, PhD, a Principal Investigator at the
Medical Artificial Intelligence and Automation (MAIA) Lab at the University of
Texas Southwestern Medical Center in Dallas (UT Southwestern).
The most significant impact of developing an
AI-based system for brachytherapy is that the new technology may allow for a
streamlined planning process, possibly superior to the manual planning process
achieved by human beings. Treatment planning of HDR brachytherapy is a highly
complex process, Dr Jia said. Several steps in the procedure, such as needle
digitization, organ segmentation, and plan optimization, require a great deal
of expertise. AI can solve these problems with greater efficiency than human
beings and with consistent quality, which could translate into clinical
benefits, he said.
In a separate study not involving brachytherapy, researchers
at UT Southwestern also investigated AI. With radiation therapy, delaying
radiation therapy by just 7 days has the potential to increase the odds of some
cancers either recurring or metastasizing.
Researchers at UT Southwestern have demonstrated AI’s ability to produce
optimal treatment plans within five-hundredths of a second after receiving
patients’ clinical data. The investigators achieved this by feeding the data
for 70 PCa patients into 4 deep-learning models. Through repetition, it was
possible to develop 3D renderings of how best to distribute the radiation in
each patient. Each model accurately predicted the treatment plans developed by
the medical team.
The investigators trained the model on 70 patients
with 5-fold cross validation and then retested it on 8 separate patients. The
hope is to automate the treatment planning process and quality assurance
process to improve treatment plan quality and planning efficiency. The
researchers said using the AI techniques, they were able to automatically
segment organs, perform a fully automated planning process, and this included
applicator digitization, radioactive source placement, and dwell time
The investigators hope to use the new AI
capabilities in clinical care after implementing a patient interface. In
addition, they are developing deep-learning tools for other purposes, including
enhanced medical imaging and image processing, automated medical procedures,
and improved disease diagnosis and treatment outcome prediction. The
researchers said all modules of the AutoBrachy system is extendable to prostate
cancer HDR brachytherapy. Upon transferring to prostate HDR brachytherapy, the
system would be applicable for HDR brachytherapy for prostate cancer, according
to the investigators.
An AI system can reduce healthy organ radiation
doses to reduce toxicity. The researchers currently are developing AI tools to
solve problems that are generally applicable to both gynecologic and prostate
cancers. For example, they have built a tool to digitize needles in volumetric
images. They found that the computation time was only less than 1 minute, much
shorter than the approximately 20 to 30 minutes for a human to manually
digitize these needles. The researchers also are developing a tool to predict
the best needle positions to guide the physicians inserting the needles. This
has the potential to reduce the number of needles required to generate a
high-quality treatment, thus reducing needle-induced trauma and improving
quality of life (QOL), according to the researchers.
This approach could be adopted easily on a
widespread basis, but an AI system for radiation treatment planning will
require more scrutiny before it could become the new standard of care. “All
machine learning technology requires initial roll out into clinical practice,
then review for performance measured by clinical outcomes and treatment
toxicity, then refinements in calculations, and further iterations. So
collectively, this process will take time too before it can be implemented as
standard of care,” said Jay D. Raman, MD, FACS, Professor and Chief, Division
of Urology, at Penn State Health Milton S. Hershey Medical Center in Hershey,
Scott Tagawa, MD, Medical Director of the
Genitourinary Oncology Program at Weill Cornell Medicine and New
York-Presbyterian, observed: “Time is only an emotional issue since weeks or
months generally don’t make much difference in time to starting therapy.
However, more accurate dosimetry or cheaper dosimetry—replacing humans with
machines—may have potential. The main caveat is that long-term follow up is
needed for both cure rates and long term toxicity rates.”
Joshua M. Lang, MD, Associate Professor of Medicine in the Carbone Cancer Center at the University of Wisconsin in Madison, said there is a great deal of excitement regarding the potential for AI to rapidly plan an optimal radiation treatment plan. “However, as with any new technology, there are multiple steps that have to be taken before we use this on patients,” Dr Lang said. “The first is demonstrating the computer program is reliable and reproducible without any errors in the code. Then, the field will have to move beyond computer modeling to actually treat patients to determine if the potential benefits actually translate into improvements in patient care.”