Updated: Jul 25, 2021
Integration of Technology and Health Care
“…AI algorithms deploy far more accurate methods when combined with analysis from radiologists, such as via pattern recognition, fuzzy logic, and neural networks, thereby improving clinical output and workflow…”
Author: Jasmine Kokkat
Artificial intelligence (AI) is a method of using algorithms to mimic human intelligence, which has contributed to cancer screening, radiology, and image analysis . AI has made advancements at accelerating rates over the past two decades. Researchers have investigated deep learning, artificial neural networks, and other ways to improve the accuracy of medical imaging . This poses the question: what positive and negative effects will AI have on healthcare in the future? Currently, AI presents some ethical and legal challenges that must be addressed before popularizing this technology in healthcare.
In what ways is artificial intelligence more efficient?
Radiologists assess medical images and report findings to detect, characterize, and monitor diseases . Hosney et al. have reported that sometimes, radiologists must analyze each image for no more than 3-4 seconds in an 8-hour workday . In this constrained time, diagnostic errors are probable, with an average error rate of 3-5% . This is equivalent to 40 million diagnostic errors annually, globally . Interpreting radiographic images relies on the education and experience of radiologists, and hence can be prone to significant human error. In fact, 60-80% of analytical errors in this type of medical imaging are caused by human perception, which includes visual memory and three-dimensional construction ability . On the contrary, AI algorithms deploy far more accurate methods when combined with analysis from radiologists, such as via pattern recognition, fuzzy logic, and neural networks, thereby improving clinical output and workflow with approximately 90% accuracy .
Artificial Neural Networks
One of the primary AI techniques currently being proposed and studied is artificial neural networks (ANN). ANNs are computational tools that were modeled from the nervous system . These tools are able to process and analyze data, then learn from these examples . So far, ANNs have been used for clinical diagnosis and image analysis in radiology and histopathology. For example, Stamey et al. developed a neural network, ProsAsure Index, to classify prostate tumours as benign or malignant with 90% accuracy . As it stands, ANNs have been tested and used to interpret radiographs, ultrasounds, CT, MRI, and radioisotope scans .
Image-Based Applications of AI
Currently, there are two different classes of AI in the field of radiology: (1) machine learning algorithms based on predefined engineered features, which have explicit parameters to characterize the shape and texture of a diseased tissue, and (2) deep learning algorithms .
Deep learning is a subset of machine learning derived from a neural network feature similar to the human brain 
The usage of deep learning in radiology has been a pivotal focus for computer and health scientists . This method can automatically learn how to categorize data into different classes without prior information from the presence of a cancerous lesion inputted by experts, for example . It allows the automatic classification of the observable characteristics of human tissues, making the diagnostic process faster for radiologists . Deep learning methods then use sample data to identify the diseased tissues .
AI is also capable of playing an important role in the diagnosis and staging of the disease . Through automation, AI can use a wide range of data and identify features with a similar procedure each time by treating them as imaging biomarkers . In this context, a biomarker is an indicator used to identify images as pathogenic . Segmentation is then done to find the extent of diseased tissue .
AI will impact the monitoring of changes over time in multiple scans for a disease-prone part of the body . While humans can identify large variations in object size and shape, it is difficult to identify the subtle variations in texture and heterogeneity within the object . Presently, this technology has been tested for thoracic, brain, abdominal and pelvic imaging, colonoscopy, mammography, and radiation oncology . Consequently, there is more research and improvements needed to make this process reliable so that it can be implemented fully in cancer diagnosis .
Usage of AI in Cancer Diagnosis
In addition to radiographic images, artificial intelligence has countless potential applications in cancer diagnosis that are yet to be developed and implemented.
Cancer is diagnosed by conducting a biopsy, which uses a microscope to examine tissue samples . As part of assessing whether the suspected tissue is malignant or not, pathologists must evaluate many factors such as cell shape, mass, and appearance . This process can yield inaccuracies due to systematic and cognitive error, like availability bias, information bias, and blind-spot bias . This has resulted in nearly one in ten cases of cancer being misdiagnosed . There are many types of cancer cells that make the diagnostic process with the naked eye more complicated.
The Shortcomings of AI
In order to overcome those obstacles, Weill Cornell Medicine and NewYork-Presbyterian developed an AI program called ‘Convolutional Neural Network, which has a similar structure to the brain . In order to “train” the program, the team exposed the network to more than 13,000 pathology images of lung, breast, and bladder cancers so that it would “learn” to identify the type of cancers in each sample . Their test was met with 92% accuracy to identify the subtype of lung cancer, and they found biomarkers for bladder cancer with 99% and breast cancer with 91% accuracy .
Artificial intelligence presents many challenges when introduced in healthcare as a means of digitalizing diagnosis. False-positive results or overlooking a diseased tissue in an image is one of the primary concerns of using AI in a clinical setting, where one error in the system can affect the diagnostic results of thousands of patients . An error in an AI system poses more ethical and legal concerns than an error a doctor may make in diagnosis, since the mistake will be repeated in the automated program . While doctors can take the route of caution by explaining the process they used to come to this result, AI presents the issue of the “black box” effect . This effect refers to the algorithm’s capability to detect abnormalities but not provide details on the process that took place to reach this conclusion .
AI technologies also present privacy concerns since the patients’ health information may be predicted by AI, revealing more to the system than intended . Hence, there is a risk of personal information being sold to banks and insurance companies. Moreover, it is crucial to have laws to avoid a violation of patient privacy. To protect this data, it is important to input information from different geographical locations and demographics so that the program will not develop a bias towards underrepresented populations .
Although AI programs have a high level of accuracy, they are highly unlikely to completely replace human pathologists in the near future. Rather, they can contribute in increasing the speed at which healthcare teams are able to analyze images since they can precisely identify complex patterns . It is critical for humans to further analyze what this data means and discuss the results with patients and their physicians. Despite underlying challenges, artificial intelligence has progressed rapidly over the years and serves as a valuable tool for humans to further develop and use in health care.
The above image illustrates a radiologist using an AI program to effectively detect breast cancer .
The figure above creatively portrays how artificial intelligence mimics the brain .
Alison MacPhee, Hadeel Alhadi, Mouayad Masalkhi, Rhea Verma
Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, et al. Artificial intelligence in healthcare: past, present and future. BMJ [Internet]. 2017 June 22 [cited 2020 Aug16];2(4):E230-43. Available from: doi:10.1136/svn-2017-000101
Waite S, Grigorian A, Alexander RG, Stephen L, Macknik SL, Carrasco M, et al. Analysis of perceptual expertise in radiology – current knowledge and a new perspective. Frontiers in Human Neuroscience [Internet]. 2019 June 25 [cited 2020 Aug 16];13(213). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6603246/
Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJ. Artificial intelligence in radiology. Nature Reviews Cancer [Internet]. 2018 Nov 30 [cited 2020 Aug 16];18(8):500-510. Available from: doi:10.1038/s41568-018-0016-5
Itri JN, Tappouni RR, McEachern RO, Pesch AJ, Patel SH [Internet]. Oak Brook (IL): RadioGraphics. Fundamentals of diagnostic error in imaging; 2018 Oct 10 [cited 2020 Aug 16]. Available: https://pubs.rsna.org/doi/10.1148/rg.2018180021
Waite S, Scott J, Gale B, Fuchs T, Kolla S, Reede D [Internet]. Leesburg (VA): American Journal of Roentgenology; 2020. Interpretive error in radiology; 2017 April [cited 2020 Aug 16]. Available: https://www.ajronline.org/doi/full/10.2214/AJR.16.16963
Bai HX, Wang R, Xiong Z, Hsieh B, Chang K, Halsey K, et al. Artificial intelligence augmentation of radiologist performance in distinguishing COVID-19 from pneumonia of other origin at chest CT. Radiological Society of North America [Internet]. 2020 Apr 27 [cited 2020 Aug 16];296:E156–E165. Available from: https://doi.org/10.1148/radiol.2020201491
Ramesh AN, Kambhampati C, Monson JR, Drew PJ. Artificial intelligence in medicine. Annals of the Royal College of Surgeons of England [Internet]. 2004 Sep [cited 2020 Aug 16]18(8):500–510. Available from: doi:10.1308/147870804290
Pena, D [Internet]. Introduction to our radiology/AI series. New York (NY): Medium; 2020. 2018 Aug 22 [cited 2020 Aug 16]. Available: https://towardsdatascience.com/introduction-to-our-radiology-ai-series-51d0e54cd3b9
Shen, D, Wu, G, Suk, H. Deep learning in medical image analysis. Annual Review of Biomedical Engineering [Internet]. 2017 June [cited 2020 Aug 16];19:221-248. Available from: https://www.annualreviews.org/doi/10.1146/annurev-bioeng-071516-044442
Xu J, Xue K, Zhangl K. Current status and future trends of clinical diagnoses via image-based deep learning. National Center for Biotechnology Information [Internet]. 2019 Oct 12 [cited 2020 Aug 16];9(25):7556-7565. Available from: doi:10.7150/thno.38065
Mayeux R. Biomarkers: Potential uses and limitations. National Center for Biotechnology Information [Internet]. 2004 April [cited 2020 Aug 16];1(2):182–188. Available from: doi:https://dx.doi.org/10.1602%2Fneurorx.1.2.182
Ahmed Z, Mohamed K, Zeeshan S, Dong X. Artificial intelligence with multi-functional machine learning platform development for better healthcare and precision medicine. Journal of Biological Databases and Curation [Internet]. 2020 Mar 17 [cited 2020 Aug 16];1-35. Available from: https://doi.org/10.1093/database/baaa010
mayoclinic.org [Internet]. Rochester (MN): Mayo Clinic, 2020. Biopsy: types of biopsy procedures used to diagnose cancer; 2020 Jan 16 [cited 2020 Aug 16]. Available: https://www.mayoclinic.org/diseases-conditions/cancer/in-depth/biopsy/art-20043922
Hajirasouliha I, Elemento O [Internet]. New York (US): Weill Cornell Medicine; 2018. Artificial intelligence aids in cancer diagnosis; 2018 Feb 9 [cited 2020 Aug 16]. Available: https://news.weill.cornell.edu/news/2018/02/artificial-intelligence-aids-in-cancer-diagnosis
Morgenstern, Justin [Internet]. Virtual: First10EM; 2020. Cognitive errors in medicine: the common errors; 2019 Sept 22 [cited 2020 Aug 16]. Available: https://first10em.com/cognitive-errors/
Shah A [Internet]. California (US): ScienceDaily; 2020. How artificial intelligence can be used to more quickly and accurately diagnose breast cancer: new paper addresses need for early and accurate tools in diagnosing cancer; 2019, July 19 [cited 2020 Aug 16]. Available: https://www.sciencedaily.com/releases/2019/07/190712151928.htm
Carter S, Rogers W, Win KT, Frazer H, Richards B, Houssami N. The ethical, legal and social implications of using artificial intelligence systems in breast cancer care. The Breast [Internet]. 2020 Feb 1 [2020 Aug 16];49,25-32. Available from: https://doi.org/10.1016/j.breast.2019.10.001
Anbil PS, Ricci MT [Internet]. Santa Monica (CA): Medical Device and Diagnostic Industry; 2020. Consider the promises and challenges of medical Image analyses using machine learning; 2020 June 02 [2020 Aug 16]. Available: https://www.mddionline.com/radiological/consider-promises-and-challenges-medical-image-analyses-using-machine-learning
Huang S, Yang J, Fong S, Zhao Q. Artificial intelligence in cancer diagnosis and prognosis: opportunities and challenges. Cancer Letters [Internet]. 2020 Feb 28 [cited 2020 Aug 16];471,61-71. Available from: https://doi.org/10.1016/j.canlet.2019.12.007
Price II NW [Internet]. Washington (DC): Brookings; 2020. Risks and remedies for artificial intelligence in healthcare; 2019 Nov 14 [cited 2020 Aug 16]. Available: https://www.brookings.edu/research/risks-and-remedies-for-artificial-intelligence-in-health-care/#:~:text=As%20developers%20create%20AI%20systems,and%20AI%20inference%2C%20and%20more.
Williams-Nyu G [Internet]. New York (NY): Futurity; 2020. AI boosts breast cancer detection accuracy; 2020 Jan 2020 [cited 2020 Aug 16]. Available: https://www.futurity.org/artificial-intelligence-breast-cancer-detection-2261322/
Kent C [Internet]. Boston (MA): Verdict Medical Devices; 2020. Healthcare AI in 2020: the ones to watch; 2020 Jan 22 [cited 2020 Aug 16]. Available: https://www.medicaldevice-network.com/features/healthcare-ai-in-2020/