Lung cancer is the most common cause of cancer death worldwide, with 1.8 million people dying from the disease in 2020. It is also the second most common type of cancer globally.
Radiation therapy, also known as radiotherapy, is a common treatment for lung cancer. More than 60% of lung cancer patients receive radiation therapy at least once.
In an effort to help improve current radiation therapy concerns, a research team from Brigham and Women’s Hospital in Massachusetts say they have developed a deep learning algorithm capable of identifying and segmenting non-small cell lung cancer (NSCLC) tumors on computed tomography (CT) scans within a few seconds.
The scientists also reported that radiation oncologists using the algorithm in simulated clinics worked 65% faster than physicians not using the algorithm.
Their study was recently published in the journal Lancet Digital Health.
According to Dr. Raymond H. Mak, a disease center leader for thoracic radiation oncology, the director of clinical innovation in the Department of Radiation Oncology at Brigham and Women’s Hospital and the Artificial Intelligence in Medicine program, and lead author of this study, radiation therapy planning is a highly manual, time-consuming, and resource-intensive process that requires highly trained physicians to target the cancerous tumors in the lungs and adjacent lymph nodes on three-dimensional images such as CT scans.
“Prior studies have shown substantial inter-clinician variation in these radiotherapy targeting tasks and there is a projected global shortage of skilled medical staff to perform these critical tasks as cancer rates increase,” he explained to Medical News Today. “Because of these quality and access gaps, we need methods that improve our efficiency and quality of tumor targeting.”
To address these issues, Mak and his team hypothesized they could train and develop artificial intelligence (AI) algorithms to automatically target cancer in the lungs and adjacent lymph nodes from CT scans used for radiation therapy planning.
“These methods, once developed, can be deployed in seconds and in different practice settings,” Mak said.
For the study, the research team used CT images from 787 people to train their AI model on how to distinguish tumors from other tissues. They then tested the AI algorithm with scans from more than 1,400 patients.
“By training the AI algorithm on segmentations of lung cancer tumors that were generated by a clinician with expertise in this task, we can theoretically replicate the skills and experience of this clinician wherever we deploy the AI algorithm,” Mak explained.
Once training was complete, researchers had eight radiation oncologists perform segmentation tasks where they identify the specific areas for treatment. The radiation oncologists were also asked to rate and edit segmentations made by another physician or the AI algorithm, without knowing which had made each segmentation.
Upon analysis, the researchers found no significant difference in performance between the segmentations made between a human and AI algorithm team, compared to those made only by human medical professionals.
The research team also found clinicians worked 65% quicker with 32% less variation when editing a segmentation created by the AI algorithm, compared to those segmentations produced manually by doctors.
Mak and his team believe there will be a direct benefit to cancer patients through thoughtful testing and implementation of human-AI collaboration in radiation therapy planning by providing patients with higher-quality tumor segmentation and accelerating times to treatment.
“Additionally, in surveys of the clinicians after they interacted with the AI, we demonstrated that the clinicians also experienced substantial benefits in reduced task time, high satisfaction, and reduced perception of task difficulty, which is an interesting additional benefit that we had not thought about initially,” he added.
Mak did point out some caveats when it comes to using AI algorithms to automate clinical processes.
“We have to ensure that human clinicians can oversee and understand the intended use and limitations of the AI algorithm,” he said.
According to Mak, the researchers identified some of the key failure modes of the AI algorithms they developed.
“For example, the algorithms had trouble targeting tumors that were present in collapsed lung tissue or next to fluid in the lung — pleural effusions — and we would need to provide warnings to end-users about these clinical contexts where the algorithms are known to perform poorly so that the clinician does not blindly accept the AI output,” he said.
“Playing this out over time, it really represents a transition from a human clinician performing a task to a human clinician acting as AI overseer and supervisor to ensure quality deployment,” Mak continued. “This transition will represent a major challenge in implementing new workflows and new clinician training in medicine.”
Medical News Today also spoke with Dr. Lisa Chaiken, a radiation oncologist and assistant professor of radiation oncology at Saint John’s Cancer Institute at Providence Saint John’s Health Center in California, about her thoughts on the new AI algorithm.
“I was happy to see how technology could help us in the medical profession be more accurate (for) better patient care,” said Chaiken, who was not involved in the study. “It’s always nice to be more efficient with your time because… that would mean I’d have more time to spend with patients. And more supportive care, which we value very much… while adding more accuracy and better delivery of care.”
Chaiken explained the process a radiation oncologist goes through to set up targeted treatment can be time-consuming and labor-intensive. She said anything that allows them to do more efficiently, accurately, and in less time will help speed up the process and free up doctors to spend more time with patients.
“And I think it would help patients get onto treatment faster, which would make them feel better or less anxious about the treatment,” she added.
Chaiken said the AI algorithm would be helpful as a double-check for doctors when finalizing targeted treatment areas.
“I think the best way to use it would be the way I often use other aids that we have,” she explained. “It should be used as an aid, not relied on completely, and it has to be checked by the physician. I think as an aid to contouring it would be very helpful.”