How artificial intelligence is improving veterinary diagnostics

Artificial intelligence (AI) has been employed in medical practice for several years, with applications sprouting up across specialties for nearly every aspect of care. One of the latest and potentially most beneficial uses of AI in medicine revolves around deep learning algorithms for diagnostic radiology. These types of AI systems are being used widely during the coronavirus disease 2019 (COVID-19) pandemic, as they have been shown to outperform senior radiologists in identifying patients with COVID-19 based on thoracic imaging results.¹

Although the value of computer-based systems in medicine has been recognized for decades, with medical diagnostic decision support systems ubiquitous by the 1990s, veterinary medicine has only recently begun utilizing AI for diagnostic purposes.^2-4 As is the case with many other areas of technological innovation, the drive for emerging AI applications in veterinary medical imaging derives largely from a desire among veterinarians and their clients for improved efficiency.

Filling the gaps

Imaging is an important component of veterinary medicine, providing essential information for routine prevention and evaluating illness and injury. Given that the demand for veterinary radiologists outstrips the ability for these radiologists to provide services, critical diagnostic results are often delayed, precluding prompt treatment.⁵ This scarcity of relevant resources also threatens to diminish the use of powerful imaging tools due to a lack of ability to interpret results in a timely manner.

While teleradiology has addressed these underutilization and inefficiency issues to some extent by enabling clinicians to share images rapidly with specialists, the ultimate value of teleradiology is still limited by the availability of those specialists.⁶ In other words, the workflow bottleneck currently exists not at the point of getting images into the hands of radiologists but rather at the point of image analysis. The most effective solutions for imaging-related gaps in care will intervene at this vital juncture, enhancing the speed with which we can process and interpret images. Deep learning applications are now providing veterinarians with precisely this capability through software systems that integrate seamlessly into clinical workflows.

Benefits of faster diagnosis

One significant change in veterinary medicine in recent years—for which there is no formal training—is the dramatic shift in consumer expectations. As McKinsey & Company reported in 2015, the sharp focus by non–health care industries on customer experience and satisfaction has altered what consumers expect from their health care providers.⁷ Rather than merely looking to us for our medical expertise, our clients now want what they perceive to be superior customer service, which includes increased speed, convenience and, when possible, cost savings. To be considered an excellent veterinarian requires that we provide our clients with rapid diagnoses that quickly clarify the health status of their pets. Importantly, there is demand not only for faster diagnoses but also for less expensive ones.⁸

With new AI applications for radiology, we can give our clients what they expect: faster answers at lower costs. While AI provides rapid insights, it can also circumvent the need for costly specialists, particularly in cases where it can help rule out complex disease or injury. Rather than having to wait on a specialist and pay specialists’ fees, we can instead incorporate AI into our practices while maintaining the ability to connect with specialists in complex cases where specialized human knowledge or opinion may add significant value. In so doing, we can improve the care we provide, save on costs for ourselves and our clients, and enhance customer satisfaction.

Augmenting specialists’ skills and improving care

Although initial AI applications in medicine proved inferior to humans in performance, newer deep learning algorithms match and even surpass human capabilities.⁶ Because deep learning algorithms can essentially view and retain enormous data sets of images, they can learn to identify patterns that may elude humans. A recent study evaluating a deep learning algorithm’s ability to distinguish between meningeal-based and intra-axial lesions in dogs using magnetic resonance imaging findings—a task often deemed challenging by veterinary radiologists— demonstrated accuracy levels of 90% and above.¹⁰ Despite their potential to help in specific contexts, these technologies are not adequate replacements for human expertise. Instead they should serve only as tools to help us overcome specific challenges in care delivery.

Identifying specific software systems that can mimic expert output in even a fraction of circumstances can drastically improve the care we provide. For example, a recent pilot study demonstrated that diagnostic accuracy via thoracic radiographs was equivalent between deep learning algorithms and board-certified veterinary radiologists in nearly 800 cases of canine left atrial enlargement.⁹ These data point to a valuable role for deep learning algorithms in veterinary radiology.

Capitalize on opportunities while avoiding pitfalls

As AI becomes more commonplace in veterinary practice, it is important that we refine the uses of these technologies so that we can employ them where they add the most value. For example, one capability that differentiates deep learning algorithms from human radiology experts is the production of quantitative, rather than qualitative, evaluations of imaging data.⁸ Considering how we can best use these quantitative data to complement our skills will be critical for ensuring that we make the most of these powerful technologies. Identifying the most innovative solutions that provide not only accuracy but also user-friendliness will empower us to put deep learning to use to provide better care.

Given that the use of these applications is already more widespread in human medicine, we should look to our human counterparts for key insights to avoid or overcome challenges related to implementing AI solutions for radiology services and how to best leverage those services.

Neil Shaw, DVM, DACVIM, is the founder and CMO of SignalPET, an innovative artificial intelligence provider whose vision is to help independent veterinarians harness advanced technology to seamlessly expand their range and depth of care. Shaw was formerly the medical founder of BluePearl Veterinary Partners and first CMO for Mars, Inc.

References

1. Mei X, Lee HC, Diao KY, et al. Artificial intelligence-enabled rapid diagnosis of patients with COVID19. Nat Med. 2020;26(8):1224-1228. doi:10.1038/s41591-020-0931-3
2. de Dombal FT, Hartley JR, Sleeman DH. A computer-assisted system for learning clinical diagnosis. Lancet. 1969;1(7586):145-148. doi:10.1016/ s0140-6736(69)91149-0
3. Miller RA. Medical diagnostic decision support systems—past, present, and future: a threaded bibliography and brief commentary. J Am Med Inform Assoc. 1994;1(1):8- 27. doi:10.1136/jamia.1994.95236141
4. Lustgarten JL, Zehnder A, Shipman W, Gancher E, Webb TL. Veterinary informatics: forging the future between veterinary medicine, human medicine, and One Health initiatives-a joint paper by the Association for Veterinary Informatics (AVI) and the CTSA One Health Alliance (COHA). JAMIA Open. 2020;3(2):306-317. doi:10.1093/jamiaopen/ooaa005
5. Cima G. Specialists in short supply: Universities, private practices struggle to find certain specialists, blame lack of residency training programs. American Veterinary Medical Association. September 26, 2018. Accessed October 15, 2020. https://www. avma.org/javma-news/2018-10-15/ specialists-short-supply
6. Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJWL. Artificial intelligence in radiology. Nat Rev Cancer. 2018;18(8):500-510. doi:10.1038/s41568-018-0016-5
7. Carrus, B, Cordina, J, Gretz, W, Neher K. Measuring the patient experience: lessons from other industries. McKinsey & Company. August 2015. Accessed October 15, 2020. https://healthcare.mckinsey. com/measuring-patient-experiencelessons-other-industries/
8. Owens JM, Lewis R, Blevins W, Silverman S, Feeney D, Mattoon J. Veterinary radiology-history, purpose, current status and future expectations. Letter. Vet Radiol Ultrasound. 2019; 60(3):358-362. doi:10.1111/vru.12713
9. 9Li S, Wang Z, Visser LC, Wisner ER, Cheng H. Pilot study: application of artificial intelligence for detecting left atrial enlargement on canine thoracic radiographs. Vet Radiol Ultrasound. 2020. doi:10.1111/ vru.12901
10. Banzato T, Bernardini M, Cherubini GB, Zotti A. A methodological approach for deep learning to distinguish between meningiomas and gliomas on canine MR-images. BMC Vet Res. 2018;14(1):317. doi:10.1186/s12917-018-1638-2