DeepHealth uses machine learning to distill lifetimes of insights from medical experts into software to assist radiologists.
Our mission is to enable the best care by providing products that clinicians and patients can trust, through rigorous science and clinical integration.
Yet making screening mammography available for all women creates a significant burden in interpretation. This strain is exacerbated by the advent of digital breast tomosynthesis (DBT; “3D mammography”), a technology that significantly boosts interpretation performance, but requires scrutinizing hundreds of image planes instead of the handful for traditional 2D mammography. We are using deep learning to create tools that will enable radiologists to interpret mammograms more efficiently and accurately, especially for DBT.
The idea of using computers to help radiologists read mammograms is not new. Despite commendable efforts (change this wording), however, traditional computer-aided detection (CAD) software has not demonstrated significant improvements in clinical accuracy, and even slows down interpretation time, given its high false positive rate and it’s recommended use as a “second look”.
Training these algorithms relies on the radiologists themselves, and other medical experts - to create gold standard interpretations to serve as learning targets.
Deep learning can achieve much higher performance levels than the feature engineering used in traditional CAD, enabling more effective use cases.
Our models have been trained on lifetimes of data and interpretations from doctors worldwide.
Beyond the data, our algorithms are informed by medical experts and are custom tailored for mammography.
Our software empowers radiologists by acting as a “smart assistant” - helping radiologists interpret more efficiently and accurately, so they can spend more time on their most challenging and important tasks.
Computers have some distinct advantages and with recent breakthroughs can be used by radiologists to delegate tasks to make radiologists more efficient, consistent and accurate.
Computers are better than humans at storing a lot of information and at doing simple calculations fast and in parallel.
Now is the time for applying computers to helping radiologists because of the unique break-throughs in algorithms, fast massively parallel hardware and storage capacity.
Our software thoroughly examines every square of every 2D and 3D imaging plane to rapidly detect calcifications and soft-tissue lesions.
Screening mammography saves lives, yet the volume of cases generated by screening puts an enormous demand on radiologists. We are using deep learning to create tools that will help radiologists interpret mammograms more accurately and efficiently, for both 2D full-field digital mammograms and 3D digital breast tomosynthesis.
Our models learn from interpretations provided by expert radiologists and pathologists to map images to gold standard diagnoses.
Our models have been trained on more data than a single clinician could possibly see in her/his lifetime.
Ultimately, our greatest goal is to improve the lives of patients, by helping detect cancers as early as possible while minimizing unnecessary callbacks.