Machine learning

Processes of digitalisation generate large amounts of data in many facets of our lives, including healthcare. Trying to make this data useful is clearly appealing. Machine learning can be seen to offer a way to use this data for the creation of systems that can automatically recognise patterns.

With machine learning it is possible to create part of a computer program based on both data and code instead of coding alone. Using machine learning algorithms we can design models to help find useful patterns in specific sets of data. To be able to find the patterns we think are useful, the model first needs to be trained with a training set of data. This dataset should be of sufficient quality and quantity, and the data has to be presented in a specific way, for instance by labeling or clustering it. This presentation allows the machine learning algorithm to adjust the parameters of the model based on the correctness of the output in order to get a higher percentage of correct outputs for the set the next time it runs. There is a balance to be struck here between the increase in correct outputs, and the capacity to generalise to unfamiliar data. To make sure the model finds the patterns we are looking for when applied to new data, a good cutoff point for the training process needs to be determined. Although the term ‘machine learning’ seems to imply that some type of machine is learning by itself, the design choices and the data used are crucial for the outcome of the process.

mHealth can both generate data that might be used for machine learning, as well as make use of diverse machine learning models. A model trained on a large set of medical images might be used in an app that helps to alert users to let a doctor take a look at certain physical irregularities on the skin, for example.1 It is also possible to use the data generated by the user themselves in order to personalise a service. For instance, by comparing sensory data that measures the amount of steps someone takes after receiving a motivational message. If someone moves more after a particular message at a particular time, an app could use this pattern to improve the timing and the content of the motivational messages.2 This mechanism would allow the type of intervention to be adapted to the response of a specific user. The various applications of machine learning will each have their technical and ethical challenges.

References

  1. X. Dai, I. Spasić, B. Meyer, S. Chapman and F. Andres, “Machine Learning on Mobile: An On-device Inference App for Skin Cancer Detection,” 2019 Fourth International Conference on Fog and Mobile Edge Computing (FMEC), 2019, pp. 301-305.
  2. Aguilera A, et al “mHealth app using machine learning to increase physical activity in diabetes and depression: clinical trial protocol for the DIAMANTE Study” BMJ Open 2020;10:e034723
Filed under: