Digital health is an area of technology that utilises computing platforms, software, and sensors for healthcare-related issues. This includes tech categories such as wearable devices, health IT, and mobile app health.
As advancements in these technological areas increase, so too does the digital integration of software and devices into life sciences and medical technologies. From this, stems many new fields in both tech and life science, such as software as a medical device.
Luke Bryan, in our R&D Tax Incentives Practice explores the potential of artificial intelligence (AI) and machine learning (ML) in medicine below.
Field of modern science
Software as a medical device is a rapidly expanding field of modern science that sees coding programmes and systems replace the functionality of many medical treatments and devices. However, with the introduction of AI to the field, are these advancements too good to be true?
The current focus on AI, heralded by the development of new AI search engines and chatbots, has led to hugely increased interest in the applications of such software in most areas, and the medical device sector is no different. Within this sector exists a unique type of medical device: medical device software. This is governed and defined by several regulatory bodies, with a generally agreed definition along the lines of “software that… is intended to be used for one or more medical purposes without being part of a hardware device.”
Essentially, this refers to software systems or applications that perform medical tasks, such as computer-aided detection, pathogen screening, in vitro diagnostics or software that allows smartphones to view clinical imaging rather than the device performing the medical function. There is already a large market for such devices, with a market size estimated at $1048 million, but the rapid growth in the market sees a projected increase in size to $10,190 million by 20281, nearly a tenfold increase in just six years.
Adding Artificial Intelligence
While these applications are already useful for physicians and patients alike, with extremely promising growth in the space, there is a desire amongst many in the medical engineering community to push the abilities of this software further with the addition of AI. There have been several applications in which AI has been adopted to improve the efficiency of medical processes, for example, for devices such as endoscopy capsules, AI guidance systems for radiology, or irregular heartbeat warning sensors.
AI can play several roles in such technologies, from an analysis tool to an active aid. For example, endoscopy capsules are ingested by the patient to take images of the gastrointestinal tract to prevent large-scale endoscopy which requires a surgery room, taking pressure off hospitals. AI becomes incredibly helpful for post-imaging analysis, where AI has been developed to flag areas of concern throughout the 24-hour or so video feed to the physician for ease of examination.
This is achieved via the deep learning process, a term used in AI which refers to showing an AI a large quantity of relevant information to establish a reference database for its tasks, in this case comparing the video feed against imagery of healthy and unhealthy digestive tracts.
This AI application can actually also analyse the images and suggest what conditions might be present in the flagged areas based on the knowledge accrued within their database, making the process even easier for physicians. This is where the issue of trust and reliability can emerge. Can the AI system accurately, consistently, and reliably identify all risk cases within the video feed, and is there a risk of influencing the decisions of the physician by suggesting what condition may be affecting the patient? The short answer is maybe.
The future of AI within medical devices
While AI systems are still developing and not every instance of a health condition will fall within two standard deviations of what the AI’s knowledge database contains, the same can be said of human physicians. There is always a risk that something might be missed during medical examination or procedure, whether AI is involved or not, but with the addition of this supplementary tool, paired with competent and vigilant physicians, there is a real opportunity in the industry to significantly decrease cases of misdiagnosis, error, and improve patient autonomy.
Of course, these improvements can only be made under the assumption that artificial intelligence within the medical device space will be utilised solely or for the most part as an assisting tool rather than a replacement for a human skilled physician, with its use applied with prudence and due diligence. This is especially important due to the “hallucination” phenomena that can sometimes occur, where AI systems generate phantom data or results and thus can provide false negative or positive results, prompting a need for careful monitoring of their results.
Conflicting results
While several studies have investigated to what extent could AI perform better than doctors, there have been conflicting results, with a study2 examining AI vs doctor diagnoses for breast cancer finding that AI made more accurate decisions than doctors, while some radiology studies3 found the doctors performed better than AI. From this, it could be said that AI performs at a comparable level to doctors, whether it be slightly better or slightly worse.
However, for now, it seems clear that AI will remain as a helpful tool to assist doctors in their work, rather than act as a form of diagnosis itself. So, can you trust the growing implementation of AI within medical devices? Well for the most part, yes, to the same extent as you trust the physician overseeing their use.
This article originally appeared in Irish Tech News and has been reproduced with their kind permission.
Get in touch
Discover more in R&D
Footnotes
- “Global Software as a Medical Device (SAMD) Market Research Report 2022 (Status and Outlook)”, Industry Research, 2022
- Kim, H.-E., et. Al., (2020). Changes in cancer detection and false-positive recall in mammography using artificial intelligence: a retrospective, multireader study. The Lancet Digital Health, [online]
- Mawatari, T., et. Al., (2020). The effect of deep convolutional neural networks on radiologists’ performance in the detection of hip fractures on digital pelvic radiographs. European Journal of Radiology, [online]