Carlos Pita1; Matías Castro2; Kyla Jones3
1BEE, Medtronic; 2 BVSC, DVM, MPhil, Medtronic; 3MSc, Medtronic
Artificial intelligence (AI) is a broad concept referring to intelligent machines or technologies capable of emulating human brain functions and imitating human abilities, including decision-making, problem-solving, reasoning, visual perception, and speech recognition. Machine learning (ML), neural networks, and deep learning (DL) are all subsets of AI.
Among the myriad applications of AI in the healthcare sector, perhaps one of the most fascinating is its direct use in patient diagnosis and treatment. Currently, AI applications in this capacity is predominately found in the medical technology space.
To illustrate the utilization of AI and identify points of convergence between technologies, pharmaceuticals, and healthcare services, two specific products were presented: the application of intelligent algorithms in insulin pumps, and the use of machine learning in the diagnosis of colon cancer. Equally important is understanding how these technologies are harnessed within the patient care pathway, either by healthcare providers or the pharmaceutical industry.
Intelligent Algorithms in Insulin Pumps: A Paradigm Shift for Diabetes Management
While diabetes management has witnessed significant advances in terms of more advanced and varied types of insulin, drugs and technology, achieving an adequate control remains elusive as the management of diabetes remains heavily dependent on patient treatment decisions, patient skills and knowledge, and the patient’s willingness to comply to treatment. This is particularly demanding for people with type 1 diabetes (T1D), who are fully dependent on insulin.
Novel algorithms are driving a profound transformation in diabetes care by automating and improving the precision of some of the decision making in the management of the condition. Specifically, advanced hybrid closed loop (AHCL) systems are capable of automatically adjusting the requirements of insulin of a person living with diabetes every five minutes based on glucose sensor data, dictated the corresponding administration or withholding of insulin. Effectively, these advanced algorithms remove the human factor from treatment decisions, thereby reducing the corresponding human error.
Foremost among the clinical benefits for patients due to these advances in technologies, is an improved time spent within the recommended range of glucose levels, otherwise known as Time In Range (TIR). A higher TIR represents a more stable glucose profile that mitigates the short- and long-term complications associated with the fluctuating patterns of glucose of a sub-optimal Diabetes management. With AHCL, up to 90% of users meet the international recommendation of 70% of TIR. This is particularly meaningful when compared to data of non-automated therapies that show outcomes ranging from 35-65% TIR, and that only ~25% of people with T1D meet recommendations.
There have been incredible advances in the software and hardware of insulin pumps over the last few years, and the algorithms are continuing to improve for the benefit of the patient.
Colon Cancer Diagnosis Using Machine Learning
Colon cancer is treatable if identified and treated early. However, early detection is challenging, with an up to 26% rate of error associated with detecting early adenomas. This challenge results in a relatively late detection of cancerous lesions, making colon cancer the 2nd most lethal type of cancer.
GI Genius, for example, uses machine learning to analyze images during a colonoscopy to identify colorectal lesions and reduce the number of missed lesions by almost half. Specifically, is a computer aided detection (CADe) tool that combines hardware and software that integrates with current colonoscopy equipment to detect polyps in real time.
The program is trained to detect polyps of various shapes and sizes and use both visual and audible alerts to highlight areas of interest to the physician, during the colonoscopy procedure. With adequate identification, the physician is able to make an immediate treatment decision.
Finally, GI Genius is built on an ever-growing data set. The technology was trained and validated with white light endoscopy videos and this data set continues to grow, from 500 images in 2017 to over 13 million images to this day. With each update, the machine learning algorithm continues to learn and incorporate improvements.
In the examples mentioned above, we see significant benefits for patients; these benefits are not achieved by replacing the physician, but by amplifying their impact though the speed, precision, and efficiency that AI can offer.
AI can be employed in medical technology software and as a tool to search for new products, among other applications within the healthcare landscape. This creates opportunities for collaboration between complementary industries, including pharmaceutical companies, medical devices, and research institutes. Moreover, it introduces new challenges for HEOR and regulators. It is at these points of convergence between diverse stakeholders that transparent discussions must continue to be fostered, promoting the responsible expansion of AI for the benefit of patients.