75% of those in charge of hospital imaging departments in the UK agree: The NHS does not currently have enough radiologists to keep patients safe. The 2018 Royal College of Radiologists Annual Workforce report, released April 2019, confirms the predicted shortfall of qualified radiologists.
It comes as no surprise, then, that according to Dr. Orest Boyko, a leading neuroradiology expert, “greater than 30% of medical imaging research now is focused on deep learning.” A 2019 study by Definitive Healthcare states that nearly one-third of imaging organizations not currently utilizing artificial intelligence indicate that they have plans to incorporate the technology within the next two years. This means that by 2022, more than half of institutions will have AI integrated into their imaging procedures.
Top implementation challenges cited by organizational reps include:
- cost (54.7%)
- strategic direction, i.e. unsure of where to start and how to implement (35.3%)
- Lack of technical expertise/personnel (33.1%)
- lack of necessary IT infrastructure (31.7%)
As with all new technologies, cost will continue to drop with time and become less of an impediment as the usefulness of the product and return on investment is affirmed by stakeholders. Dr. Boyko acknowledges that in order for the radiology community to embrace new algorithm-based technologies, word-of-mouth communication and validation is also important in addition to clinical studies confirming the accuracy of AI diagnoses.
An Assistant to Humans, Not a Replacement
For those practitioners and stakeholders who fear that AI technologies will suddenly replace jobs, the term “artificial intelligence” was coined in1956. Although much more visible in recent years, AI technologies have been a long time in the making.
AI can streamline radiologists’ clinical workflow by “highlighting for us high-probability positive cases with potentially critical findings that should be read first,” says Dr. Boyko. Instead of organizing the workload chronologically, AI’s deep learning capabilities can determine which images are more likely to contain conditions requiring treatment priority. This “heads up” can help improve communications between radiologists and the referring ordering doctor.
Another useful application of machine learning technology is natural language processing (NLP), in which a neural net can be trained to organize and sort data contained in anonymized patient records. This tool is much more efficient than searching for data terms manually within the electronic records.
Integrated AI and related technology is here to stay. Don’t be afraid—learn what it can do to help in your healthcare setting.