CAD systems were once widely used in the US. Thirteen years ago, 74% of mammography screening studies were read with CAD assistance. Those were the first computer-based attempts to detect and highlight visible structures in medical images. Over the last 5 years a new generation of AI emerged. While AI and CAD both aim at assisting radiologists in the analysis and interpretation of medical images, they are not the same. What makes them different? Let’s take a deep dive to understand.
Definitions
Computer-Aided Detection (CAD) is a diagnostic software that is based on traditional machine learning techniques. Machine learning employs algorithms to analyze information, learn from it and make informed decisions based on the knowledge gained. Compared to modern AI, traditional CAD uses more limited techniques that can only be trained on small datasets and is not capable of characterizing findings accurately. It is not capable of improving its performance independently by learning techniques by itself.
AI2 tool (Artificial Intelligence Aided Interpretation tool) is a new generation of AI software using deep learning techniques. Deep learning organizes algorithms in layers to create convolution neural networks. These networks are trained to recognize complex patterns, learn and make decisions on their own. In comparison to traditional algorithms, AI algorithms can use very large datasets and continue to improve as data available increases, achieving human-like performance.
Accuracy
Studies evaluating CAD’s performance showed that it did not offer any significant benefits to radiologists. On the contrary, CAD negatively affected radiologists’ accuracy and productivity. Its use led to an increase in false-positive rate, which resulted in more needless patient recalls and biopsies, causing additional anxiety in patients.
On the other hand, AI-based software helps radiologists become more efficient. Capable of highlighting small lesions that the human eye might not notice due to fatigue or other reasons, when used in mammography, AI helps improve breast cancer detection, lower false-positive rate, and can detect cancers earlier. AI provides more reliable results. Moreover, thanks to the diversity of data used to train AI, it can detect rare types of lesions that radiologists with less experience may have never seen before. In addition to that, AI evolves over time and augments its diagnostic quality.
Reading time
Traditional CAD systems were designed to detect structures in an image. The goal was to ensure the radiologist did not overlook any visible structures when analyzing mammograms. CAD displayed loads of irrelevant data pinpointing all types of lesions, without distinguishing malignant and benign findings. Radiologists had to spend more time reviewing all these findings, which greatly slowed them down. Thus, a study revealed that a mammography interpretation assisted with CAD increased reading time by approximately 25 percent.
Meanwhile, innovative AI technologies, such as MammoScreen®, were created not only to detect suspicious findings, but also to characterize them and display only the meaningful ones. This lets radiologists concentrate their attention on where it is needed most. It can help physicians read confidently, thus faster, benign cases and focus on more complex cases.
User experience
The latest AI-based software provides radiologists with a better user experience. For instance, in addition to highlighting findings, software like MammoScreen® assigns a global level of suspicion, helping radiologists make a faster decision: to recall or not. These tools behave like a second reader on the physician’s side.
With known accurate performances, AI software can not only be used at the end of the reading protocol but can be used at the beginning to directly focus on lesions of concerns to spend the most time on what is valuable.
AI-based software offers wider workflow enhancement possibilities.
Most AI are cloud-based, which ensures you always have the latest AI version at your practice and allows lightweight deployment.
Integration in practice
Traditional CAD is known as on-premise software; it requires the installation of cumbersome equipment on each workstation and the on-sight presence of an engineer. Conversely, latest generations of AI systems are based on cloud technologies that require light on-site installation. These installations can be supported remotely, which provides for a decrease in integration expenses and time, and also provides the possibility of deployment during a pandemic situation such as COVID-19.
MammoScreen is a new generation AI2 software, providing radiologists with decision-making support for breast cancer screening. If you would like to know more about how it works, we invite you to sign up for a free demo and see for yourself how you can benefit from it.