Excellence in Software Engineering
Artificial Intelligence, The Hesperus of Healthcare
14 Juli 2021

Authors: İsmail EL – SW Architect / Dr. Deniz KATIRCIOĞLU ÖZTÜRK – Group Manager / Healthcare Programs

Performing some certain tasks in an automated fashion requires the development of complex algorithms which can be used in such a sophisticated way that researchers and clinical practitioners can benefit from them to investigate, hypothesize and figure out complicated medical problems. Based upon this, the application areas of Artificial Intelligence (AI) in healthcare are versatile. At the highest level, assistance in diagnostics and prognostics, drug discoveries, genome studies, clinical trials, and improving therapeutics are only few of the current applications of AI in healthcare.

 

Market Growth of AI in Healthcare

The use of AI in healthcare refers to the analysis and interpretation of huge amounts of data sets so as to leverage better decision making, effective management of patient data, and eventually creation of a personalized approach to ensure the medical excellence.

First, the use of AI in recognizing healthcare parameters and complications, without a doubt, would help make better decisions faster than the human brain does, under pressing time constraints.

Second, AI provides a great deal of added value to the otherwise unhinged “information management” for all the actors in the point-of-care including the patients. With the means of telemedicine employed, patients, now, can access to the most needed healthcare services at the most appropriate time and price. On the professional front, health caregivers can also increase their abilities so as to deepen their medical know-how through AI-driven content and support their continuum in medical education.

Third, with the ability to analyze large amounts of “-omics” data, AI is able to identify patient-specific, or rather genome-specific treatment options and molecule interactions. As well, all the pertinent AI methodology is situated as an advantage for the pharma companies as they usually struggle against the expensive and the lengthy process of molecule and drug discovery. Similarly, the use of AI and high computational power in searching for patterns to design novel or alternative treatment methods would save the stakeholders from the burden of costly and low-populated clinical trials that are usually wrapped in legal sanctions. This trend would eventually lessen the dependency on humans and present a plenty amount of scope for what AI can create.

Growth opportunities are likely to be ragarded as the end of rainbow by companies unless significant investment is done to harness the power of AI in healthcare. However, the main opportunity for billion dollars growth in healthcare AI has long burgeoned, that is not unexpected, at all. So much so that AI fully meets the needs of the healthcare industry – such a perfect match in heaven – the major ones are as follows:

  • Robot-Assisted Surgery
  • Diagnosis Support
  • Virtual Nursing Assistants
  • Administrative Workflow Assistance
  • Fraud Detection
  • Automated Dose Dispensing
  • Connected Machines (IoMT)

 

Known Limitations of AI in Healthcare

Despite its proclaimed potential, there exist a number of already addressed limitations concerning AI.

Initial Adoption

The teething problems experienced with the introduction of any kind of new technology is nothing new, but must be coped with when it comes to large scale adoption and compliance to existing regulations. Success stories need to be emphasized and presented for future encouragement. But still, for these success stories to come true, it requires some early adopters of healthcare sector to put a stake in the ground by kickstart the process in one way or another.

Data Privacy

Privacy within healthcare, as a matter of course, is highly sensitive, the systems and infrastructures should be preserved against malicious attacks and the data privacy must be ensured.  On top of the conventional measures taken, AI itself brings brand new approaches in distinguishing among breaches and ensuring data security.

Black-Box Difficulty

AI usually does not provide any answer to the question ”why?”. Therefore, since the reasoning behind the clinical decisions are not supported by ownself of the technology, guesswork is required more often than not to justify how the decision was taken. This lack of reasoning and justification may imply a degree of unreliability towards the technology and fail to relieve the burden from the shoulders of the healthcare professionals. However, thanks to the recent advances in eXplainable AI (xAI) approach, this doubt over the reliability of AI may be eliminated in the very near future.

Stakeholder Resistance

When it comes to the adoption of AI in healthcare, nearly all the stakeholders including point-of-care professionals, patients, insurance and pharmaceutical companies, hospitals etc. tend to hold on to the conventional ways of delivering the practice and acquiring the services. Resistance to keep pace with the technology at any micro level may likely to  lead to potential failures herewith to the incorporation of the technology in macro level.

 

AI-based Healthcare Applications in ICterra

Despite the limitations and drawbacks encountered as is with each new technological development, new advancements relevant to AI are announced in healthcare almost every day.

ICterra’s share of effort on AI is mostly framed around the fields of Radiology and Oncology. At this very stage, the abundance of clinical, imaging data and quantification with radiomics diversify the problems to be focused on and approaches to be adopted [1]. In particular, analysis and smart algorithms to be built on screening data are somewhat immune from the tornado of elements that bring critical comments to the AI ​​perspective. When it comes to medical screening, one of the prominent radiological modality that comes to mind is mammography (nowadays, operated digitally as Full Field Digital Mammography (FFDM)).

It is now an undisputed fact that breast cancer is the most common type of cancer in women and, unfortunately, it ranks first among cancer-related mortality in women [2]. According to authorities such as American Cancer Society and American College of Radiology, women aged 40-44 should have the option to begin their breast cancer screening with mammograms, while women above the age of 45 should get them annually, even if they are deemed as asymptomatic. Considering on this scale, the operational burden of the work based on incidence is quite tremendous. And from a data scientific point of view, this extent would accumulate into high volumes of mammograms and make the problem in question eligible to be handled with the popular “deep learning” paradigm, especially as the images are labelled quite diligently to provide evidence to critical interventions and oncology medications [3]. It should also be be noted that; a significant number of cases may be missed or misdiagnosed due to morphological variation and low signal-to-noise ratio [4].

Based on these critical points, ICterra has taken the scientific and technical mission of developing a solution that would relieve the workforce based on Radiology expertise under usual and extraordinary (eg. covid-19 pandemic) care routines and would aim to reduce the possible false positive / negative rates by increasing the sensitivity of anomaly detection in mammograms. The underlying methodology relies on state of the art image processing and deep learning approaches for mammographies and is based on a grand-scale dataset aggregated in collaboration with one of the prominent radiology departments in national academia. Therefore, this project is part of the data-science pipeline for a realistic use of AI-enabled clinical decision support in ICterra with an understanding that endeavours are by no means complete but rather an absolute work in progress.

 

References

[1] Mao, N., Yin, P., Wang, Q., Liu, M., Dong, J., Zhang, X., … & Hong, N. (2019). Added value of radiomics on mammography for breast cancer diagnosis: a feasibility study. Journal of the American College of Radiology16(4), 485-491.

[2] Azamjah, N., Soltan-Zadeh, Y., & Zayeri, F. (2019). Global trend of breast cancer mortality rate: a 25-year study. Asian Pacific journal of cancer prevention: APJCP20(7), 2015.

[3] McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., … & Etemadi, M. (2020). International evaluation of an AI system for breast cancer screening. Nature, 577(7788), 89-94.

[4] Hubbard, R. A., Kerlikowske, K., Flowers, C. I., Yankaskas, B. C., Zhu, W., & Miglioretti, D. L. (2011). Cumulative probability of false-positive recall or biopsy recommendation after 10 years of screening mammography: a cohort study. Annals of internal medicine, 155(8), 481-492.

Frühere Artikel

Using ChatGPT for Rest Endpoint Testing

Using ChatGPT for Rest Endpoint Testing

Today, UI test automation is one of the most popular topics encountered during test automation discussions. In order to accomplish any UI test automation task effectively, framework usage is strongly suggested. When building up a test automation framework, variety of models and tools can be used.

Navigation