Para fornecer o conteúdo mais relevante aos leitores do HealthCare Business News, pedimos que você compartilhe um pouco de informação sobre quem você é (leva dois segundos e pronto).
Registo
RadNET's DeepHealth AI facilitates breast cancer detection a year sooner than radiologists
DeepHealth's AI algorithms may be able to detect breast cancer one or two years earlier than standard practice
DeepHealth, a subsidiary of RadNET, is touting a new study in which its AI algorithms detected breast cancer a year or more in advance of current practice.
The software surpassed five full-time, breast-fellowship-trained expert radiologists in reading the same mammogram screenings, with the findings suggesting that the algorithms may help detect cancer one to two years earlier than standard interpretation in many cases.
“Most cancer researchers believe that patients whose cancer is detected and treated before it has metastasized have greater survival and can avoid more expensive therapies, and so detecting cancer one or two years earlier, when it has a lower chance of metastasis, implies both higher quality and lower cost of care,” lead author Bill Lotter, chief technology officer and co-founder of DeepHealth, told HCB News.
The study also lauds its ability to interpret mammograms accurately with a lack of annotated human data available. Such information must show the status of the cancer and is essential for continuously improving an AI solution’s ability to detect cancer. This information is often difficult to obtain, and even with it, the solution must still be tested across different clinical sites and patient populations in different scenarios.
DeepHealth’s approach copies how humans often learn by progressively training the AI models on more difficult tasks. This enables them to learn in each successive training stage and detect cancers more accurately while relying less on highly annotated data. The approach and validation include 3D mammography, which is growing in adoption due to its clinical effectiveness. Its large data size, however, makes it a challenge to assess with AI applications.
“We tailored our AI approach to address the 'needle-in-a-haystack' nature of mammography, where lesions tend to be relatively small and subtle. Any similar scenario where a physician must comb through a significant amount of data to identity anomalies is appropriate for our techniques. One such example would be prostate MR for cancer detection.”
DeepHealth recently submitted its first AI-based product, a software program that prioritizes 2D and 3D mammograms. The solution is based on the same core technology described in the study, and is expected to be approved in 2021.