Blog

Experts Weigh in On the Evolving Age of Artificial Intelligence in Medical Diagnostics

A new article in Clinical Microbiology Newsletter reviews the current role of emergent Artificial Intelligence (AI) and Machine Learning (ML) technologies in microbiology diagnostic testing. The authors of the article examined the increasingly prevalent use of Artificial Intelligence (AI) in our day-to-day lives and explored the role of AI in digital image analysis of parasitology samples, gram stains, and bacterial culture. The article also reviews the state of AI technologies in fields of MALDI-TOF mass spectral data analysis and genome sequencing.

AI is simply defined as programmed algorithms or sets of rules that software uses to make decisions very much as a human would. ML is a powerful application of AI that allows systems to automatically learn and improve from data and training without being explicitly programmed. There are many examples of AI and ML in everyday life; AI decisions power ridesharing and navigation apps. AI automatically filters spam emails, drives search engine & video streaming recommendations, and even goes as far as curating content on our social media accounts.

AI and ML are increasingly being used in medical diagnostics and are exceptionally good at analyses that conform well to complex decision tree-based interpretation. Over time, ML allows the AI to transform into  “adaptive AI” evolving as results are observed and recorded. This method is particularly useful for complex diagnostic data analysis when there is too much data, too many variables, or measurements that are too detailed to be easily conceptualized by the human mind. AI can give equal consideration to massive amounts of data, far more than the human brain can concentrate on.

“Identifying Relevant Data Points”

A key hurdle the AI must overcome in these data-rich contexts is the ability to rule out data that is not important or relevant to the task at hand. To address this issue, many recently developed AI software platforms use larger data sets that compare to recognize the changing variables and classify the relevant data. Another powerful tool reviewed in the article is the use of  ML in conjunction with deep learning algorithm models called Convolutional Neural Networks (CNN). CNN uses many interconnected algorithms, or neural networks, to analyze data and make decisions, much like the connections the human brain uses to recognize patterns and make decisions. The article mainly focuses on the use of AI and ML in the context of clinical microbiology, where the technologies have already impacted workflows in many medical laboratories.

The authors of the article report that AI is currently being adapted to assist in the interpretation of Gram Staining. A recent study utilized Inception Version 3, Google’s CNN designed for object detection in Google images. The scientists in the study trained the CNN to identify microorganism morphology on over 100,000 gram slides, achieving a 95% whole-slide classification accuracy. Although the use of artificial intelligence in gram staining is still being developed, the authors of this article see great potential in AI’s ability as a tool that works alongside technicians for accurate and efficient gram stain analysis.

Likewise, the authors of the article also explored the use of AI in parasitology, another field that requires extensive microscopic examination by highly trained laboratory staff. In malaria diagnostics, technicians rely on identification, specification, and quantitation based on the visual morphological characteristics of the parasites found in patient specimens. One major diagnostic challenge in parasitology is that in many regions impacted hardest by parasitic diseases don’t have the staffing, instrumentation, or infrastructure required for analysis of large volumes of patient specimens; many of which are negatives.

The potential advantage of using AI in parasitology is the improved efficiency and accuracy in settings where a full staff of highly specialized personnel is not realistic. The authors raise the potential of using commonplace electronic devices like smartphones to capture images and data in hard to reach areas.  This data then can then be processed by AI and flagged if abnormalities are observed. The use of AI is continually being developed for these purposes in the field of parasitology and shows great promise in helping diagnose parasitic conditions regions that are typically underserved.

AI and automation have already been making impacts on the clinical microbiology space as well. Automated blood cultures, susceptibility testing, and molecular platforms are commonplace in numerous laboratories around the world. In clinical bacteriology, many laboratories are seeing greater efficiency using automated upfront specimen processing platforms like COPAN’s WASP® and Digital Plate Imaging platforms like WASPlab®. Further, the rise of Digital Plate Reading (DPR), like COPAN’s PhenoMatrix™ AI, has been making significant improvements to laboratory efficiency in many settings.

COPAN’s  PhenoMatrix™ AI software is being used in laboratories to segregate “growth”, “no growth”, “mixed growth”, or even singular pathogens based on the laboratory-defined thresholds. In chromogenic agars, DPR is already in use in several laboratories to detect the presence or absence of the target organisms.

Importantly, the average time to result reporting for urine cultures was significantly decreased by almost 5 hours for negative specimens and by 3.5 hours for positive specimens. This decrease in time to result is achieved by enabling the AI software to interpret the images as soon as they are captured instead of waiting for a human to be ready to read the images. The authors of the article were quick to point out that these efficiencies are vital because the clinical microbiology industry is experiencing a shortage of experienced staff to interpret plated specimens. Further, this shortage of skilled laboratory staff will likely be intensified over the coming years as a large number of laboratory personnel are set to retire by the year 2022.

As AI use in bacterial culture interpretation becomes more sophisticated, it is expected that the software, powered by ML, will be capable of further species identification in complex specimens where multiple pathogens are present. The increasing complexity of the software will likely further the efficiencies that laboratories are experiencing by using AI-driven DPR to decrease the staff hours required to result cultures, improving turnaround time, and positively impacting patient outcomes.

The article further discusses the use of AI and ML in the MALDI-TOF analysis and whole-genome sequencing of bacterial samples. Both fields can use AI and ML to learn and characterize minute differences between specimens to identify patterns that emerge from the differential data analysis. These patterns found by the software can be used to find strain-specific bacterial susceptibilities and resistances. Currently, the AI and ML software for these methods are still being refined and typically take longer than conventional culture-based antimicrobial susceptibility testing. However, new breakthrough instrumentation and software will likely close the gap in terms of efficiency in the near term.

In conclusion, the use of Artificial Intelligence along with Machine Learning and Deep Learning Neural Networks are already being used and further developed for medical diagnostics. The authors of the article state that “It feels inevitable that the number of AI tools, the quality and reliability of the analyses performed by AI software, and the integration of AI into the clinical microbiology laboratory workflow will grow in the coming years and decades” [1]. As the full potential of these systems is still yet to be completely realized, it is all but certain that the field of medical diagnostics will be profoundly changed by the coming wave of technological innovation. Read the full article here.

Click Here to learn more about COPAN’s Automated Microbiology & Artificial Intelligence Platforms

References:

  1. Kenneth P. Smith, Hannah Wang, Thomas J.S. Durant, Blaine A. Mathison, Susan E. Sharp, James E. Kirby, S. Wesley Long, Daniel D. Rhoads, Applications of Artificial Intelligence in Clinical Microbiology Diagnostic Testing, Clinical Microbiology Newsletter, Volume 42, Issue 8, 2020, Pages 61-70, ISSN 0196-4399, https://doi.org/10.1016/j.clinmicnews.2020.03.006.