Discover how computer-aided detection (CAD) systems are continuously evolving
In mammography screening, the workload for radiologists is extremely high – and continues to grow. With digital breast tomosynthesis, the data volume will increase even further and, consequently, the reading time. But there are smart tools available that support radiologists in saving on reading time.
Common tools that help reduce reading time
There are already various tools widely in use that support radiologists in decreasing the evaluation time:
- Slabbing reduces the amount of images to be viewed, as e.g. 1 mm slices are joined to form thicker slices.
- Correlation links define the area where it’s possible to find the lesion when switching between several projections of 2D and 3D images.
- O’clock positioning helps the reader automatically and reproducibly define the exact lesion position.
- Automated measurement tools help define the location of a lesion by immediately displaying the distance to the skin line, nipple, and chest wall.
Computer-aided detection in mammography
Since being introduced to the market in the late 1990s, the CAD algorithms for mammography have been constantly evolving. The first algorithms used to detect conspicuous tissue in 2D mammograms were laboriously programmed – and not highly reliable. Nowadays, CAD systems utilize image analysis and deep learning technologies1). Information from thousands of mammography images are incorporated into the algorithms, enabling the product to distinguish between characteristics of cancerous and normal tissue.2) These CAD systems clearly identify lesion candidates in both 2D mammograms and 3D tomosynthesis volumes.
Reading workflow with a CAD system
Whether for 2D or 3D data, CAD systems are mostly used as part of a second reading or as general reading support, depending on national regulations. They support radiologists on more than one level. By interpreting the images and marking suspicious tissue, they may improve the detection of cancer in the breast and alert radiologists to areas that need further analysis. They also facilitate locating lesions and microcalcifications – and thus, reduce reading times.
In a 3D reading workflow, for example, when clicking on a lesion in the 2D image, the user is directly led to the specific 3D slice where the lesion is best visible. In that way, clinicians can easily and quickly find even small lesions again in subsequent examinations like an ultrasound scan for dense breasts.
This is what a CAD workflow in screening might look like:
CAD is also controversially discussed
There are various studies available that take a closer look at the efficacy of CAD. Adding CAD to single reading usually shows an increase in sensitivity and/or cancer detection rate, while there’s no significant difference when comparing double reading with single reading plus CAD. Many studies also report an increase in recall rate when adding CAD.3)