Prostate cancer is the second most common cancer in men in the United States (after skin cancer) and is the second leading cause of cancer deaths (after lung cancer). The American Cancer Society predicts that approximately By 2024, 35,000 American men will die from prostate cancer. So anything that can help doctors detect and treat prostate cancer more quickly could save thousands of lives.
A crucial tool for an accurate diagnosis of prostate cancer is MRI scanning. A high-quality MRI scan can help surgeons detect the presence of a tumor, the extent of the tumor within the gland, and the presence of any spread to adjacent organs.
However, MRI scans still need to be interpreted by human radiologists. And people can make mistakes. In particular, recent research has shown that human radiologists are It seems more likely that a scan later in the working day is interpreted as ‘malignant’when ‘fatigue, workload and circadian rhythms can influence decision-making.’
This is no surprise. Other researchers have shown that “radiologists are more prone to errors after 10 hours of clinical interpretation.” However, the study performed by Dr. Anton Becker and colleagues were able to quantify this with regard to the specific issue of prostate cancer diagnosis. (Dr. Becker’s article also quoted the same thing The ‘hungry judge’ effectin which the sentences of judges on the parole boards were said to be more severe as they became more tired and hungry during the workday, and to become more lenient after they had been rested and fed.)
Prostate cancer awareness month concept. Banner template with light blue ribbon. Vector … [+]
Dr. Becker and colleagues also suggest that artificial intelligence can help reduce error rates due to human fatigue. Fortunately, AI algorithms are reaching the point where this could soon become a practical reality.
Researchers from the Mayo Clinic and Massachusetts General Hospital have succeeded in developing a deep learning AI system in accuracy almost equivalent to that of expert human radiologists. Furthermore, a combination of humans and AI performed better than either one alone.
Another team of researchers from UCLA was able to use an AI system to… exact degree of tumor spread in a way that “45 times more accurate and consistentthan conventional methods. Being able to more accurately determine the boundary between tumor cells and normal tissues can be extremely useful in treatment planning, allowing doctors to make optimal use of “surgery, focal therapy, radiation therapy, hormone therapy, chemotherapy or a combination of these treatments“to destroy or remove the cancerous tissues while causing as little damage as possible to nearby healthy tissues.
As with other current developments in medical AI, these tools cannot (yet) completely replace human doctors. Dr. Takahashi from the Mayo Clinic team commented, “I don’t think we can use this model as a standalone diagnostic tool.” Instead, he sees the AI “as an addition to our decision-making process.” And as always, these research results need to be validated in larger groups of patients before they are ready for widespread daily clinical use.
Nevertheless, I am very happy to hear about such fruitful collaborations between physicians from leading academic medical centers and AI scientists. The coming years of the AI revolution should bring more good news like this for patients and doctors alike.