Amy Morris looks into how applying machine learning tools to mobile health applications could work in the future...
Machine learning has huge potential to transform the world of healthcare, and we have already started to see just how.
In this article, I will briefly discuss how mobile apps that employ sophisticated machine learning models can significantly improve health outcomes, the potential risks that need to be considered, and how instead of threatening to make doctors obsolete, AI can serve as a useful asset in your doctor’s toolbox.
Medicine is entering the fourth industrial revolution: Artificial Intelligence.
AI already plays an integral role in our day-today lives; you wake up and unlock your phone with facial recognition technology to check the time, you then scroll through your personalised social media feed, and you even use AI-powered travel apps to find the best route into the office. So why not employ AI to take charge of your health?
With AI-powered wearable devices like smartwatches predicted to have 224.27 million users by the end of 2023 , enormous amounts of data from vital signs and other health metrics can be analysed in real-time to detect changes from normal patterns and potentially alert healthcare professionals when intervention might be necessary.
(42% of smartwatch users have discussed the data from their watches with their doctors )
An example of this would be the healthcare technology company AliveCor with their portable ECG device known as KardiaMobile which can be attached to a smartphone or tablet and can take ECG readings at a high clinical standard to detect arrhythmias and alert the user of any irregularities all whilst being neatly wrapped up in a friendly user interface downloadable from the App Store .
The aim of technology such as this is to empower patients to be more proactive in their healthcare, making it easier to educate individuals on their condition and bridge the gap between patients and healthcare providers.
Arguably, one of the most exciting ways AI can be used to make your doctor’s job easier is through algorithms designed to detect early signs of disease. Healthcare professionals train for several years, becoming superhuman pattern recognition experts in their respective fields. AI systems can also learn to recognise patterns within healthcare data, relying on vast amounts of data to learn from.
So by training AI algorithms to make sense of variable raw patient data; such as medical images and dermatology photos could be ground-breaking in assisting healthcare professionals in identifying potential abnormalities sooner.
Doctors can see thousands of patients and go through tens of thousands of hours of training to condition their brains to recognise slight abnormalities in radiological images, cytology slides, and skin lesions. Despite their expert level detection, doctors are simply human and don’t have the time or powers to see every pixel to the resolution and detail as specialised AI-powered programs.
In a study from 2019 published by Tschandl et al in The Lancet Oncology Journal on the effectiveness of an AI-powered skin lesion analysis platform called DermEngine found that a group of medical professionals predicted malignancy of lesions with 95.8% accuracy whilst the AI algorithm had an accuracy of 96.3%.
So how does it work?
The first important step to utilising AI systems for use in healthcare is data collection. We are living in a data-driven world with 2.5 quintillion bytes of data created every day in 2021 and healthcare is no different . There are many various sources for data in healthcare such as medical records, images, patient histories, or genetic information. Making sense and processing all this data is a huge task.
Record keeping for an individual could contain many inconsistencies and noise from their raw data, so this data needs to be structured in a way that can be analysed by AI, example by normalising data formats or removing duplicate records.
The next step would be extracting features. This process essentially involves transforming features from the data into a format that the AI model can understand in order to identify relevant patterns. Actually training the models involves feeding them labelled data and providing the desired outcomes. For example, labelling an X-ray of a patient with pneumonia. The model will begin to recognise particular visual features in the image and correlate this finding with the diagnosis of pneumonia.
We are on the cusp of a new industrial revolution when it comes to using AI in healthcare, but with a sector relying on the use of highly sensitive and extensive personal data, ensuring privacy and security is a significant worry many people have seeing as 79% of UK healthcare providers reported at least 1 data breach in the last 2 years .
Another concern with utilising AI algorithms is the presence of bias. AI systems can produce unfair results or bias which they inherit from the dataset used in their training, this could impact underrepresented populations. For example, many dermatological diseases present differently on dark skin, but many AI models are trained to spot abnormalities mostly found on light skin. This could increase rates of false negative results for skin abnormalities on darker skin and risk missing diagnoses .
Whilst it’s exciting and slightly terrifying to imagine putting such responsibility in 1s and 0s it’s important to note that healthcare professionals will always play a crucial part in validating any recommendations made by an AI model and that they are most likely to be used as a tool to aid healthcare practitioners.
Ultimately the potential benefits of AI in healthcare applications are immense but it’s vital that we approach with caution and consider the potential harms by employing robust regulations, continuing to thoroughly research the field as well as addressing ethical considerations so that AI technologies can be responsibly and safely used to improve patient outcomes now and in the future.
Written by Amy Morris
 Daniel Ruby. Smartwatch Statistics 2023: How Many People Use Smartwatches. DemandSage website. [Internet]. Published 2023; Available from: https://www.demandsage.com/smartwatchstatistics/. Accessed September 26, 2023.
 Jacqueline DeMarco. Nearly 70% of Americans Would Wear a Fitness Tracker/Smartwatch for Discounted Health Insurance. ValuePenguin website. [Internet]. Published 2022; Available from: https://www.valuepenguin.com/fitness-tracker-smartwatch-health-survey. Accessed September 26, 2023.
 Al-Alusi, M.A., Ding, E., McManus, D.D. et al. Wearing Your Heart on Your Sleeve: the Future of Cardiac Rhythm Monitoring. Curr Cardiol Rep 21, 158 (2019). https://doi.org/10.1007/s11886-019- 1223-8
 Tschandl P, Codella N, Akay BN, Argenziano G, Braun RP, Cabo H, et al. Comparison of the accuracy of human readers versus machine-learning algorithms for pigmented skin lesion classification: an open, web-based, international, diagnostic study. Lancet Oncol. 2019;20(7). doi:10.1016/S1470-2045(19)30333-X.
 Aditya Rayaprolu. How Much Data Is Created Every Day in 2023? techjury website. [Internet]. Published 2023; Available from: https://techjury.net/blog/how-much-data-is-created-every-day/. Accessed September 26, 2023.
 SOTI. THE TECHNOLOGY LIFELINE: CHARTING DIGITAL PROGRESS IN HEALTHCARE. SOTI website. [Internet]. Published 2023; Available from: https://soti.net/media/mcbn5b1b/sotiindustry-report-the-technology-lifeline-charting-digital-progress-in-healthcare-letter.pdf? utm_source=press-release&utm_medium=organic&utm_campaign=publicrelations&utm_content=2023-05-pressrelease-healthcarereport-uk. Accessed September 26, 2023.
 Daneshjou R, Vodrahalli K, Novoa RA, et al. Disparities in dermatology AI performance on a diverse, curated clinical image set. Sci Adv. 2022;8(32):eabq6147. doi:10.1126/sciadv.abq6147.