Robot at the Barbican Centre, London. Photo: © Tim Sandle
To gain an insight into the likely technology changes, the top five trend predictions for 2023 have been compiled by BeeKeeperAI (which develops confidential computing for healthcare artificial intelligence).
AI is an important area for healthcare, given that industry predictions suggest AI algorithms will improve patient outcomes by as much as 40 percent while also reducing treatment costs by up to 50 percent (as per Frost and Sullivan data).
The key factors driving the acceleration of healthcare AI in 2023 are identified as:
Trend #1 − Technical improvements in methods for improving data security and protecting privacy
One concern with collecting health data is with patient confidentiality. In recent years, the process to anonymise the data has become more robust. These developments in privacy-preserving capabilities like confidential computing have democratized access to the technologies making them much easier to implement broadly.
Trend #2 − Growing demand for real-world clinical data
One concern with healthcare AI is the ability of a system to adapt to new scenarios. Therefore, it is important that following an algorithms’ validation, developers must be able to account for model drift as the algorithm encounters new data which can potentially adversely impact performance and accuracy.
Trend #3 − Addressing biases in algorithms
It is important that AI addresses the issue of bias so that the appropriate treatments are given. This includes accounting for gender and ethnic variations. Algorithm debiasing approaches are an important tool for reducing potential bias in healthcare algorithms, however, as things currently stand, more research is needed to identify the most effective approaches for different settings (so-called ‘algorithmovigilance’).
Trend #4 − Mandated data sharing
It is important that researchers share data and many research institutes and governments mandate this. However, shared data must be protected and hence sharing encrypted data is the way forwards. This enables data providers and algorithm developers to share resources securely while minimizing the risk of exposure for both the data and the algorithm.
Trend #5 − The need to cut costs.
As AI becomes more established, the pressure grows for costs to be lowered. This includes the necessity to deliver validated, generalizable algorithms to assist in identifying which patients would benefit from which treatment at a specific moment in time.
