MERLIN ENGELKE, MS, data scientist at the Institute of Artificial Intelligence, PhD candidates at the University of Duisburg-Essen, and the principal investigators in “Refine and International Validation of Machine Learning Algorithms for Classifying Acute Leukemia Subtypes using routine laboratory data,” explores the research. American Journal of Managed Care® In front of him This Sunday, a presentation at the European Haematology Conference in Milan.
Abstract presents an international validation of previously studied machine learning algorithms, an artificial intelligence (AI) tool designed to classify acute lymphocytic leukemia subtypes (acute lymphocytic leukemia). [ALL]acute myeloid leukemia [AML]and acute promyelocytic leukemia [APL]) uses routine laboratory data from over 5,500 patients in 14 countries.
This transcript was lightly edited. The caption was automatically generated.
Transcripts
Why is a timely diagnosis of acute leukemia so important? Why is this particularly challenging in resource limit settings?
Well, it's essential to diagnose quickly, especially with highly aggressive leukemia subtypes. [die] Eventually. It is also very important, especially in lower or mid-term countries, as they may not have all the equipment to make a diagnosis.
What have previous studies shown about the potential of machine learning models to improve early detection of acute leukemia?
There were more early research [about] Imaging, and now we do it with tabular data, so it's completely different. It is promised to show progress towards better decisions.
Building on that foundation, what was the main purpose of your research? What methods have you used to investigate this?
This study was an improvement on existing research. First of all, I wanted to find out if a claim was made. In fact, it got worse. This was a problem and then I made some improvements. Split the data into two.
First of all, it was trained [on] Although only for adult patients, outlier detection was superior to previously calculated cutoffs, so refined it. We essentially reorganized the model in our cohort and for the pediatric cohort.
Can you explain the important findings in detail? Was there anything particularly noticeable or surprising?
The insight was that we could have a generalized model, but I don't think we're at some point [where] It can be said that it can only rely on AI-based model decisions. However, it can give you the direction that is heading towards the leukemia subtype.
Does AI analyze to detect acute leukemia subtypes?
It really depends on the subtype we are talking about. For example, AML saw platelets better, whereas APL, plasma, and leukocytes were more important. So here it really depends on the subclass.
Can you also explain in detail how outlier detection tools addressed the limitations of the algorithm, especially when predictions were less reliable or diverse across sites?
We weren't really distributed on some sites regarding data points for adults in outlier detection. What we did there was that we received over 90% of results. [the] Model trust for independent test data.
Based on your findings, how can this AI tool be integrated into clinical practice, especially in resource-constrained environments?
Essentially, it's just an xgboost model, so you should be able to host it online on your website. It's not that resource intensive, so you can place it on a web server, make it public and make sure your data is not stored for the long term. Even if the centre has a really strict GDPR [General Data Protection Regulation] For example, you can host the rules yourself as well.
In hematology, AI is really important and I think it will play a very important role in the future now, but there are not many lectures now. [on it] As you can imagine.
