Table 2.

AI functionalities related to the perception and reasoning

AI functionality	The supporting role in the radiology workflow
Segmentation (8%*)	Segmentation is a functionality of many applications that designate a specific organ. The segmentation not only liberates the radiologists from this task, but also optimizes the limited attentional resources that radiologists have during the work and potentially reduces both false-positive and false-negative errors by supporting them to focus on the most relevant part and the image. This is sometimes achieved through suppressing less relevant aspects of the image to reduce the information overload and thus focuses the attention of radiologists on the more important aspects of the image.
Quantification and extraction of features (28%)	Many of the applications quantify certain aspects of the image (e.g., bone density), measure some aspects of the organ (e.g., brain volume), or extract quantitative features from the image (e.g., the level of coronary calcium scores). The outcome is often presented as numbers and charts (when it is done on a series of images).
Detecting and highlighting the suspicious areas (42%)	This category of functionalities focuses on a particular pathology or abnormality and looks for their signs (e.g., nodules, strokes, and high-density tissues) and highlights them. These applications are often trained for a particular (common) disease and aim to ensure the accurate examination of the images and to help radiologists detect and decide about certain problems.
Comparison, cross-referencing, and longitudinal analysis (8%)	As one step further to provide medical insights, some applications compare the different images of one patient to detect the changes in certain aspects over time (e.g., tumor size). This function is sometimes used to find other similar cases that have been previously diagnosed and therefore providing insights into further consultations and comparisons by the radiologist.
Diagnosis and classifying abnormalities (11%)	Diagnosis is a common functionality that builds on the previous functionalities, but combine them with a judgment regarding the likelihood of certain problems. This judgment can be achieved by comparing with the normal/healthy standards (e.g., the normal brain size), as well as identifying certain problematic areas (e.g., broken bones or tumors). These applications vary depending on how much they frame their outputs as “the actual diagnosis” or as “pre-diagnosis” to be further examined by the doctors.
Prognosis (2%)	Only a few applications offer the possibility of predicting the likelihood of certain diseases or problems based on the inspection of the current examinations. This prognosis is often focused on a particular problem and sometimes uses additional clinical information, next to the information that is extracted from the image.
Patients profiling and synopsis, and case prioritization (1%)	A group of applications actually do not work directly with the images, rather extract additional information related to the patient from the previous reports and electronic medical records, next to each patient’s image (e.g., as dashboard or personalized view about a patient). The analytical insights that they offer enable radiologists to have a broader overview about the patient’s history and conditions and therefore more accurately examine their images.

*Percentages reflect the share of applications having this functionality