Type:
Educational Exhibit
Keywords:
Artificial Intelligence, CT, MR, Computer Applications-Detection, diagnosis, Image verification
Authors:
R. Schlegel; Melbourne, VIC/AU
DOI:
10.26044/ranzcr2021/R-0009
Imaging findings OR Procedure details
The key findings of this review indicate that advances in machine learning techniques have the potential to improve all aspects of the radiology workflow. This includes improvements in the accuracy, safety, efficiency and productivity of clinical decision making, as well as providing better patient care through improved detection and interpretation of radiological findings and reporting.
For instance, machine learning is capable of extracting regions of interest on the basis of expert knowledge in cancer characterisation such as tumour volume, shape, texture, intensity and location. The most robust features are then selected and fed into machine learning classifiers. Conversely, deep learning simultaneously performs feature extraction, selection and ultimate classification across several layers during the training process. As layers learn increasingly higher-level features, earlier layers might learn abstract shapes such as lines and shadows, while other deeper layers might learn entire organs or objects. These processes may be applied throughout all areas of radiological imaging in order to support clinicians across a broad range of disease diagnoses (see examples below). These include areas such as colitis classification, cancer metastasis, and trauma imaging (e.g. identification of pneumothorax).