Artificial intelligence (AI) is starting to play a major role in medicine – and particularly in diagnostics. This is an exciting time in the industry, with many predicting that AI will revolutionize the way things are done. Simply put, AI is helping us to see data in ways we couldn’t see before. It’s like putting the available diagnostic data under a microscope, which allows us to analyze it with more confidence and make better data-driven decisions.1
Why AI? How Artificial Intelligence Can Improve Healthcare
“By automating tasks and processing more data, AI will empower HCPs to find more meaningful solutions to both basic and complicated issues in healthcare.
AI tools have the potential to advance patient care, disease management, prevention and treatment, at an unprecedented level.” – IQVIA, 2019
By providing better insights into the collected data and making more sensitive technologies a reality, artificial intelligence (AI) can enable better detection of diseases and conditions. This empowers healthcare providers to make more informed decisions, leading to better outcomes for patients.1
There’s so much to say about AI and IVD testing, it’s hard to condense it all into one blog post. The examples here are just a few of the exciting developments in this field.
Some compelling examples are available in point-of-care (PoC) testing and oncology.
AI and Point-Of-Care Testing
PoC testing is where clinical laboratory testing is done close to the site of patient care (e.g. pharmacies, doctors’ rooms, clinics, community centers or at home), facilitating rapid turnaround time.3,4 It’s also used in under-resourced situations, such as in rural areas with poor access to healthcare facilities.1,3
Here are a few recent examples of how the use of AI in in vitro diagnostics (IVD) is improving PoC testing:
Lateral flow immunoassays (LFIAs) have been combined with AI to test for HIV. LFIAs are often used in PoC testing, but the accuracy of test-result interpretation ranges from 80% to 97%. The test lines that appear are typically very faint and, if the tester doesn’t have the right training and experience, the results may be misinterpreted.3
|Why are accurate results so crucial? Read more about the dangers of a false positive result.
A deep-learning algorithm has been developed to counter the challenges that come with LFIA testing. Patients were tested using PoC tests and then images of the samples were captured with a high-quality Samsung tablet camera integrated with an android camera application. The results were sent to a mobile health system.3 A field study of the algorithm showed higher sensitivity (97.8%) and specificity (100%) than those that were visually interpreted by humans, including experienced nurses.
This means that highly accurate HIV testing can be done in a PoC setting, simply by using a tablet in combination with this algorithm.3
Anemia and sickle cell disease have high mortality and morbidity rates in resource-limited countries. Recently, a PoC test was developed to help with the diagnosis of these blood disorders.3
The PoC microchip electrophoresis device can measure anemia and hemoglobin (Hb) variants using an artificial neural network (ANN)-based machine learning algorithm.3 Using a blood sample, the device takes 10 minutes to separate the Hb into subtypes.
The sample is diluted in a standard calibrator and electrophoresed on a cellulose acetate paper. The device measures the Hb concentration by comparing the Hb band intensity with that of the standard calibrator band.3 High-resolution images of both bands are captured, including all relevant pixel information.
Each frame in the image video is split into two separate channels, and this information is fed to a trained ANN that examines the intensity ratio pattern and reports the Hb concentration. These results are used to determine whether the patient has anemia.3
This AI system has been shown to detect anemia with 100% sensitivity and 92.3% specificity. It also detects sickle cell disease with 100% sensitivity and specificity. This highlights how AI can enable testing in a PoC setting for multiple conditions using a single device.3
Tropical Disease Testing
AI can be used together with bright-field microscopy testing to diagnose dangerous tropical diseases like malaria and schistosomiasis.
In patients with clinical malaria, microscopy testing alone is associated with only 75% sensitivity. It’s even more difficult to accurately diagnose patients with low parasite levels or partial immunity. It takes a properly trained and experienced healthcare worker to make the diagnosis, and this training may be lacking in a resource-limited healthcare setting.3 An automated microscopy system has been developed to address these challenges.
EasyScan Go captures images and uses a convolutional neural network (CNN) algorithm to interpret the images and make a diagnosis. A CNN is a machine learning model based on the human visual system; it uses mathematical operation of convolution to interpret images.3
The sensitivity of EasyScan Go was 91.1% and specificity was 75.6%. Specificity increased to 85.1% when using high-quality slides. Although there’s still room for improvement, the use of AI is expected to significantly improve the detection of malaria.3
Schistosomiasis is a neglected tropical disease, caused by a flatworm, which affects around 236.6 million people, mostly in Africa. It’s diagnosed using brightfield microscopy to identify the presence of flatworm eggs in a urine sample. It can be easy to miss milder infections and low numbers of eggs, so it’s important that the tester has the right experience but, as mentioned, experienced testers aren’t always available in under-resourced settings.3
A low-cost schistoscope has been designed to help detect schistosomiasis. It’s a high-quality digital microscope and slide scanner, constructed using easily accessible parts. It uses a CNN model to identify the images, referring to a data bank of 5 000 flatworm egg images captured from collected urine samples. A preliminary study showed that the schistoscope has 80% sensitivity.3
AI can be harnessed to improve the quality of life of patients in rural and remote areas and can help enable IVD testing in parts of the world where it’s previously been inaccessible.1,3
AI and Oncology
AI also has the potential to significantly impact cancer research, diagnosis and treatment.1,6
Predictive models powered by AI can determine a person’s likelihood of getting a certain cancer by identifying the risk factors. This means they can be closely monitored and quickly treated if needed. Early detection and treatment can significantly reduce mortality.6
Deep learning (DL) has already been used to predict cancer risk from mammogram images. MIT (Massachusetts Institute of Technology) has developed a DL model that correctly identified 30% of future breast cancer patients as belonging to a high-risk group. This is compared to 18% identified by doctors using traditional methods.6
AI-based cancer classifiers, which are trained using thousands of cancer samples, can recognize patterns that are too subtle for the human eye to notice. This helps doctors to prescribe more targeted cancer therapies, with better outcomes for the patient.6
KIST (Korea Institute of Science and Technology) has developed a cancer diagnosis technology that uses tactile-neuron devices with ANN-based learning methods. When pressure is applied to a potentially cancerous site on the patient’s body, the electrical spikes generated increase or decrease depending on the stiffness of the object encountered. This helps diagnosticians to determine if the site is a tumor or not.6
Personalized treatment planning is increasingly important for cancer diagnosis and therapy to ensure better outcomes. AI has helped in the development of treatment choices over the past 50 years, using imaging technologies that allow for:6
- Early detection
- Highly personalized radiation therapy
- Targeted chemotherapies
MIT researchers now predict that sensor technology and the Internet of Things (IoT) will be able to gather extensive personal data from patients, which can be used to create individualized treatment programs. AI and machine learning are able to evaluate huge amounts of data and select the most effective course of action.
|Using the patient’s individual biology rather than population biology can boost the effectiveness of treatment by providing improved precision medicine.6
The Challenges Facing AI in Medicine
While AI is a very powerful tool, we do need to consider how it could go wrong.1 AI faces constraints because of concerns around data ownership, security and privacy, data bias and the non-availability of labelled data.6
|“We should always remember that AI can and will make faulty decisions in some situations because its decision making is predictive and probabilistic in nature. As such, there are no regulations or guidelines to establish who is legally accountable when AI malfunctions or causes harm in the course of providing a service.” – Sebastian AM, 2022.6
Human medical experts are still needed to interpret results and provide a human-verification element to the process. This means that AI isn’t expected to replace medical professionals, but rather to help them make more effective treatment decisions.6
The Future of AI in Healthcare
As AI is advancing so rapidly, it can be hard to accurately predict where it will take us next.1
Anthony Austin, Global Marketing Manager at Medix Biochemica, provides a fitting analogy: When the computer was first invented, we knew it was going to revolutionize the way we did things, but we couldn’t predict exactly how. Today, technology like video calling has become an everyday occurrence – a reality that would have seemed unthinkable a few years ago. The same principle applies to AI: we know it’ll change things dramatically, but the exciting part is we still don’t know exactly what that will look like.1
How Medix Biochemica is Adapting to the Rise of AI
The use of AI won’t likely impact the need for products like antibodies and antigens. However, we may need to make these technologies more sensitive and specific, or improve other performance parameters that will enable assay platforms (and AI algorithms) to pick up smaller details and produce more robust results. Medix Biochemica seeks to be the most trusted partner to the IVD industry, continuously improving our products to meet the growing demands of our customers.
The need for biospecimens will also remain unchanged, as testers require them to prove that their technologies will work on real patient samples. By providing these products and ensuring a high level of quality, Medix Biochemica can empower customers embracing the power of AI to develop their tests with confidence.1
What sets Medix Biochemica apart?1
- Adaptability: If a customer is using new techniques that require higher purity or specificity, we can adapt to meet those requirements.
- Agility: We're agile and able to shift our focus to meet the customer’s needs. We enable you to do what you need to do and we grow with you.
- Customer service: We build close relationships with our customers to meet your needs more effectively.
- A comprehensive portfolio: There are many players in the field who offer a few quality IVD products, but Medix Biochemica provides a combination of quality and quantity. Our portfolio includes a wide array of antibodies, antigens, enzymes and biospecimens.
The use of AI is an exciting, challenging and often unpredictable trend, both in IVD and in medicine in general. We look forward to enabling our customers in their work as developments continue to unfold.
- Expert opinion. Interview with Anthony Austin, Global Marketing Manager at Medix Biochemica. September 29, 2023.
- Artificial intelligence and in vitro diagnostics: advancing patient care. IQVIA. Accessed October 5, 2023. https://www.iqvia.com/blogs/2019/03/artificial-intelligence-and-in-vitro-diagnostics-advancing-patient-care.
- Khan AI, Khan M, Khan R. Artificial intelligence in point-of-care testing. Ann Lab Med. 2023;43(5):401-407. doi:10.3343/alm.2023.43.5.401.
- Larkins MC, Thombare A. Point-of-care testing. In: StatPearls. StatPearls Publishing; 2023. Accessed October 5, 2023. http://www.ncbi.nlm.nih.gov/books/NBK592387/.
- McRae MP, Rajsri KS, Alcorn TM, et al. Smart diagnostics: combining artificial intelligence and in vitro diagnostics. Sensors. 2022;22(17):6355. doi:10.3390/s22176355.
- Sebastian AM, Peter D. Artificial intelligence in cancer research: trends, challenges and future directions. Life (Basel). 2022;12(12):1991. doi:10.3390/life12121991.