Stay up thus far with free updates
Simply register for Artificial intelligence myFT Digest – delivered straight to your inbox.
A brand new model based on artificial intelligence can accurately detect several sorts of cancer, assess treatments and predict survival rates, the most recent advance in medical diagnostics based on this technology.
The model, often known as “Chief,” is a breakthrough due to the wide selection of tumors it might probably analyze and its ability to predict treatment outcomes for patients, say its inventors at Harvard Medical School.
Chief highlights how AI has contributed to improvements in image-based diagnostic techniques, partially because it might probably detect the importance of features that even an experienced human eye might miss.
“Our goal was to develop a versatile, versatile ChatGPT-like AI platform that may perform a wide selection of cancer assessment tasks,” said Kun-Hsing Yu, assistant professor of biomedical informatics on the Blavatnik Institute at Harvard Medical School. “Our model proved to be very useful for several tasks related to cancer detection, prognosis and treatment response in numerous cancer types.”
While recent breakthroughs in artificial intelligence have led to fears of misuse of the technology, optimists argue that it also has the potential to bring long-term advantages to humanity in areas reminiscent of medicine and climate science.
Chief, described in a paper published today in Nature, works by reading digital slides of tumor tissue. It was trained on 15 million unlabeled image sections after which on 60,000 whole slides of tissue covering 19 different cancer types.
The idea was to be certain that Chief could relate detailed changes in a single area of tissue to its larger context, the researchers said. They tested its performance on nearly 20,000 whole-body images from 24 hospitals and patient cohorts all over the world.
Chief outperformed other AI diagnostic methods by as much as 36 percent in detecting cancer cells, predicting patient outcomes, and identifying tumor origins and the presence of genetic patterns related to response to treatment, the paper said. Unlike another current models, it’s so versatile that it maintains its performance whatever the techniques used to acquire and digitize the tumor cells, they added.
Chief showed an overall accuracy of nearly 94 percent in cancer detection, reaching 96 percent for esophageal, stomach, colon and prostate tumors. The ability to link tumor cell patterns to specific genomic abnormalities could help recommend the perfect treatments without the necessity for expensive and slow DNA sequencing, the scientists said.
The model provides further insightful information concerning the tissue surrounding the tumors, reminiscent of the presence of a bigger variety of immune cells in cancer survivors who’ve survived long-term cancer in comparison with those that died earlier, the study says.
If Chief and similar approaches prove valid in further research, they may very well be used to “discover cancer patients early who may benefit from experimental treatments targeting specific molecular variations,” even in countries where this just isn’t currently happening, Yu said.
Because of their speed and talent to acknowledge patterns, AI models are proving to be an increasingly useful ally for medical imaging professionals. While not yet perfect, they may be helpful in triage, as a second opinion, or to realize insights that a physician can have missed or be unaware of.
Chief appears to be a crucial recent pan-cancer tool in a growing field of diagnostic fundamental AI models, said Professor Eric Topol, founder and director of the Scripps Research Translational Institute in California.
In April, researchers from Harvard Medical School at Boston's Brigham and Women's Hospital announced two models — often known as Uni and Conch — to read, interpret and classify microscopic slides of patient tissue. They performed well in diagnostic tasks starting from disease detection to organ transplant assessment, and in addition showed some ability to discover recent and rare diseases.
The emerging recent diagnostic models based on artificial intelligence promise to “provide extraordinary insights from whole slide images, including improved accuracy of diagnosis and prognosis,” Topol said.
