By Randy Jones

Artificial intelligence (AI) is helping reduce cost, patient suffering, and potentially saving lives. One example of this is AI that assists radiologists in the interpretation of Prostate MR imaging, which will be discussed further in this article. Another goal of the article is to help reduce the confusion in the radiological space regarding what AI software actually is. This confusion is largely due to software manufacturers claiming that their products use AI when in reality many of these use simplistic decision algorithms to provide convenience and potentially improved throughput to the user. Artificial intelligence is defined as a computer system able to perform tasks that normally require human intelligence. This includes systems as simple as a decision flowchart for automatically displaying patient data, to complex model systems that automatically detect anomalies in radiographic images. Just as radiologists require extensive training and practice to accurately perceive anomalies in images, so does an AI algorithm performing the same task. The software program must be able to learn from huge quantities of sample data with known truths (proven pathologies) to measure performance against – hence, the term machine learning. In contrast, relatively simple programs are sufficient to arrange and display images, for example. Though these programs can mimic human decisions, the models behind them need not be particularly sophisticated.

There has been a proliferation of radiological software products across all modalities and on a broad range of organs—MSK, lungs, and so on—employing various levels of AI in the past 6-8 years and too numerous to list here; however, this link provides a slightly dated listing of companies and products from the perspective of those submitted for CE marking in the EU; hence, not a complete list by any means. Given the recent advances in Prostate MRI, we’ll use this topic to compare and contrast the various forms of AI algorithms and their outputs.
Artificial intelligence > machine learning > deep learning
First, we’ll clarify the terms Machine Learning and Deep Learning. Machine Learning is a subcategory of AI and includes algorithms that “learn” as they are presented with more data. These algorithms utilize statistical, probabilistic, and optimization techniques so that the best decisions are learned from the data without the need of explicit programming. Random forest is an example of a machine learning algorithm. Deep Learning is a subcategory of machine learning and includes multilayered models constructed from artificial neural networks. These algorithms can learn complex associations from data to inform decisions, much like how the human brain works.

Artificial Intelligence Homepage