Computer-aided detection (CADe) has made significant inroads into U.S. clinical practice. Its key application in preventive mammography screenings has been to mark conspicuous structures and sections that require further investigation. Less than 5% of screening mammograms were interpreted with CADe in the United States in 2001, three-quarters of screening mammograms in the Medicare population were interpreted with CADe by 2008 following the Food and Drug Administration’s (FDA) approval of the first CAD systems for mammography screening in 1998 and the Centers for Medicare and Medicaid (CMS) reimbursement for its use in 2002. The Breast Cancer Surveillance Consortium database reports that 83% of screening mammograms in 2012 were interpreted with CADe. The current CMS reimbursement for CADe is roughly $7 per exam and many private insurers pay over $20 per exam for CADe. The FDA, which estimates 40 million mammograms are performed annually, defines CADe as:
Computerized systems that incorporate pattern recognition and data analysis capabilities (i.e., combine values, measurements, or features extracted from the patient radiological data) intended to identify, mark, highlight, or in any other manner direct attention to portions of an image, or aspects of radiology device data, that may reveal abnormalities during interpretation of patient radiology images or patient radiology device data by the intended user (i.e., a physician or other healthcare professional).
Early Detection Is Becoming Faster and More Accurate
Breast cancer is the most common cause of cancer mortality among women in developing countries, and the second cause of cancer death in more developed regions. Early detection in order to improve breast cancer outcomes and survival remains the cornerstone of breast cancer control globally, and mammography screening remains an effective method for early detection. In the United States, mammography has helped reduce breast cancer mortality by nearly 40% since 1990. Double reading, which is the standard practice in mammography screening programs in developed regions, has been shown to increase the cancer detection rate by 5% to 15% in Europe. The shift from film to digital imaging technology has also contributed to improved cancer detection rates through improving the visualization of smaller lesions and greater detection of calcifications. In the United States more than in Europe, however, it has also contributed to an increase in the abnormal interpretation rate.
The first generation of CADe products in the late 1990s relied on machine learning techniques, such as shallow neural networks and support vector machines, which relied on clinical domain knowledge to support manual classification techniques. Furthermore, the performance of these algorithms was limited by underlying rules and statistical models. The high number of false-positives generated by early CADe systems, for example, has been attributed to this analytical approach. The increasing use of deep learning techniques and, in particular, convolutional neural networks, involves algorithms that are now trained using large data sets. By feeding algorithms with radiologist-annotated images and a “ground truth,” the system automatically learns about the image features rather than being programmed on what to look for. Deep learning methods typically produce faster and more accurate results, particularly for soft-tissue analysis.
Imaging Modality Advancements Will Improve Digital Mammography
Deep learning systems can be trained to identify features using large datasets, so the algorithm development times are massively reduced compared with the traditional approach of “manually crafting” algorithms. In mammography screening, where CADe’s application has been largely restricted to two-dimensional (2D) imaging, iCAD’s SecondLook and Hologic’s ImageChecker are examples of leading CADe solutions for breast cancer screening based on 2D images. A recent development focus among vendors has been to extend the utility of CADe to newer three-dimensional (3D) imaging modalities, particularly tomosynthesis. In breast tomosysnthesis, deep learning-based CADe products can help increase sensitivity in detecting cancer and significantly improve workflow by reducing the reading time associated with 3D images compared to 2D mammograms. iCAD’s PowerLook Tomo Detection, which extracts areas of interest from the 3D planes and blends them onto a synthetic 2D image, has been developed with deep learning technology. Based on initial trials, the company claims that the reading time associated with breast tomosynthesis over 2D mammography is reduced by an average of 29.2%, with no change in radiologist performance.
Building on these advances in digital mammography, Tractica expects to see a wider selection of CADe solutions emerge for use with different imaging modalities (e.g., breast magnetic resonance imaging (MRI) and breast ultrasound). This has the potential to improve upon the performance of earlier generation technologies.