A team of AI and medical researchers at startup Ataraxis AI, who are also affiliated with a host of institutions across the U.S., has announced the development of an AI model that they claim is more accurate at predicting the progression rate of breast cancer than standard tests now administered in hospitals.

The group has published a paper describing their model and outlining how well it has performed during testing on the arXiv preprint server.

Prior research has shown that cancers develop at different rates, even cancers of the same type. Some breast cancers progress quickly, for example, while others grow slowly. This difference makes it difficult for health care providers to develop the optimal treatment regimen for a given patient.

So researchers have developed tests such as Oncotype DX to gauge the aggressiveness of a given cancer in a patient—such tests may be genetically based or rely on tracking progression over a short period of time.

In this new effort, the team at Ataraxis AI, which revealed its presence publicly just last year, has developed a machine-learning AI model to estimate the likely rate of progression for a given patient.

To train their model, the team partnered with several hospitals, giving them access to large databases of tumor progression imagery along with patient statistical information. They also note that they have improved the accuracy of their model by developing several models that work differently but offer the same types of results and then averaging their predictions. The approach, they claim, results in removing errors.

Thus far, the team has tested their model on 3,500 patients using historical data. The team then compared their estimates of progression rate risk with results of standard tests such as Oncotype DX and found their model to be up to 30% more accurate.

The team plans to continue their research, looking to improve their model's accuracy rate—they also plan to look into developing other tools aimed at helping doctors more accurately assess breast cancer characteristics. They are planning to make software using their AI model available to health care facilities as early as next year.