7 things to look for when purchasing radiology AI software


Artificial intelligence (AI) has a great deal to offer the field of radiology. Medical imaging applications can handle big data and facilitate lesion detection, image acquisition, segmentation and interpretation. Radiology AI can support, enhance and corroborate the work of skilled radiologists, relieving pressure on staff and resources and providing better outcomes for patients. However, to ensure AI solutions meet the needs of radiology departments, they must be critically assessed. 

Evaluating commercial radiology AI software packages to find one that is suitable is a highly complex task. With this in mind, we’ve created a quick seven-step guide to the key questions radiologists should ask before choosing an AI radiology solution. Hopefully, this can help guide discussions between stakeholders and manufacturers and ultimately, make it easier to identify the right tool.

1. How does the software deal with bias, erroneous results or malfunctions? Consider questions such as: 
  • Does the model adapt according to the availability of new data? Or is performance based on data provided only at the training stage? 
  • How does it incorporate user feedback (for example when radiologists detect errors)? 
  • How was the AI model trained and what steps were taken to eliminate bias? 
  • What ongoing support does the manufacturer provide to resolve issues? 
2. Does the radiology AI software adhere to data protection regulations? 

There is an abundance of AI software on the market, however not all AI models are created equal. You should consider whether any software you purchase supports your legal obligation to data protection and privacy laws. 

3. Is the AI model understandable and explainable?  

Is it possible to interpret and explain results, e.g. through visualisation? Can non-radiologists (including physicians and patients) understand the data?

4. Is the software interoperable with other systems?  

Interoperability is an important consideration, particularly with regard to systems that are already in use in the department/hospital.  

5. Can the IT infrastructure of the radiology department support the package? Or will new hardware need to be purchased? 

Discussing this with your IT department sooner rather than later is important. Associated costs and time implications need to be taken into consideration.

6. How does the package impact workflow in the department? 

For example, through the production of automatic reports, or the prioritisation of cases. With this in mind, you should consider whether it will impact the allocation of resources, including human resources. Will changes need to be made to the departmental reporting structure to ensure the new system runs smoothly? 

7. How good a fit is the radiology AI software within your department? 

Consider subjective factors that relate to the running of your department on a micro level, including: 

  • Budgetary constraints 
  • Technological abilities of staff members (and their willingness to adopt new systems)  
  • Ease of use and intuitiveness of the system 
  • Compatibility with existing processes, organisational culture and workflow in the department 

Introducing new software can be somewhat daunting, but as the healthcare sector becomes more digitalised, it’s inevitable this will become the norm. On top of this, the input of radiologists in the decision-making process is crucial.  New tech should meet the needs of the department, reduce workloads and relieve time pressures on existing staff members, and ultimately, improve patient outcomes.  

Zegami provides end-to-end imaging AI consultancy, with unique image visualisation and analysis tools and machine learning expertise. Our aim is to enable you to easily analyse and understand large curated visual data sets, whilst improving the efficiency and accuracy of your workflows. If you’d like to find out more about how we can help your radiology imaging AI projects, get in touch