With 1.8 million deaths from the illness in 2020, lung cancer is the most prevalent type of cancer mortality globally. Additionally, it is the second most prevalent kind of cancer worldwide. Lung cancer is typically treated with radiation treatment, also referred to as radiotherapy. Radiotherapy is administered to more than sixty percent of people with lung cancer at least once.
A research group from Brigham and Women’s Hospital in Massachusetts claims to have created an intense learning algorithm capable of recognizing and sectioning NSCLC (non-small cell lung cancer) on CT scans (computed tomography) in order to address prevailing radiotherapy issues. The experts also stated that in modelled clinics, radiation oncologists who used the algorithm performed their work 65% quicker than those who did not. Their research was a new release in the Lancet Digital Health journal.
Overcoming obstacles in radiation therapy
Dr. Raymond H. Mak, the test’s main researcher and director of medical technology development in the department of radiation oncology at Brigham and Women’s Hospital as well as the AI in Medicine program, claims that organizing radiation therapy is an extremely manual, resource-intensive, and time-consuming procedure that needs highly skilled healthcare professionals to approach the malignant cells in the lungs and nearby lymph nodes on 3D imaging techniques like CT scans.
He told that previous research has found significant inter-clinician diversity in these radiation targeting activities and there is an estimated global scarcity of competent medical workers to carry out these crucial jobs as cancer rates climb and that they need strategies that enhance the effectiveness and standard of tumor treatment due to these access and quality gaps.
Mak and his group theorized that they could teach and create artificial intelligence (AI) algorithms to autonomously pinpoint cancer in the lungs and surrounding nodules from computed tomography scans used for radiotherapy treatment in order to solve these problems.
To train its AI model to recognize cancer from different tissues for the research, the study group analysed CT scans from 787 individuals. After that, they used images from over 1,400 patients to evaluate the AI program. “By training the AI system on lung cancer tumor divisions created by a healthcare professional with experience in this activity, we may theoretically reproduce this clinician’s skills and knowledge wherever we implement the AI algorithm,” Mak added.
8 radiation oncologists were trained before performing segmentation duties, which involved determining the precise treatment zones. Additionally, without disclosing who had created each segment, the radiation oncologists were asked to score and alter segmentations created by either a different doctor or an AI computer.
Analysis revealed no discernible performance difference between segmentations created by a human and AI algorithm collaboration and those created only by human medical specialists. In addition, the study group discovered that when revising a segmentation generated by the AI system as opposed to segmentations performed physically by doctors, clinicians performed 65% faster and with 32% less inconsistency.
Pros and cons
By carefully evaluating and implementing human-AI cooperation in radiotherapy planning, Mak and his colleagues expect that cancer patients will directly benefit from faster treatment times and better-quality cancer segmentation. In survey results of the clinicians following their interactions with the AI, he continued, “we also showed that the clinicians encountered significant benefits in limited project duration, greater satisfaction, and lowered impression of task complexity, which is a fascinating additional advantage that we did not initially believed.”
Mak did make a few cautions on the use of AI algorithms to digitize medical procedures. He stated, “We must make sure that human therapists can supervise and comprehend the intended usage and restrictions of the AI system.”