SimulaMet researchers recognized with ACM Health Best Paper Award
Published:
In 2020, an extensive study on machine learning (ML) models for gastrointestinal (GI) tract disease classification was published in the ACM Transactions on Computing for Healthcare. This week, the researchers received the ACM Health 2020 Best Paper Award for their outstanding contribution.
The paper, titled An Extensive Study on Cross-Dataset Bias and Evaluation Metrics Interpretation for Machine Learning Applied to Gastrointestinal Tract Abnormality Classification, is written by Vajira Thambawita, Debesh Jha, Hugo Lewi Hammer, Håvard D. Johansen, Dag Johansen, Pål Halvorsen and Michael Riegler.
“The recognition of this paper as the top paper for 2020 highlights the impact and potential of our research in this field,” says Vajira Thambawita.
About the award
The ACM Best Paper Awards was established in 2024 to recognize research with the most significant contributions and impacts to computing for healthcare. Papers from 2020 to 2024 have been reviewed by the inaugural committee to recognize the best paper among all papers published in the journal that year. Learn more about the awards on their website.
The study underscores the importance of cross-dataset evaluations—testing models on datasets other than those used for training—to better simulate real-world conditions and improve model robustness across diverse hospital settings.
To identify the generalizability of their ML models, different datasets for training and testing were used in the evaluations. The results demonstrate the challenge of actually building a generalizable model that could be used across different hospitals.
The authors raise an important question for future research: are generalizable ML models for medical applications possible?
In this regard, the authors have since contributed to enhancing the generalizability of machine learning in analyzing GI tracts by introducing real datasets and proposing methods for synthesizing GI tract data to address data limitations. Selected key follow-up scientific papers are listed as follows:
- https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0267976
- https://www.mdpi.com/2075-4418/11/12/2183
- https://www.nature.com/articles/s41597-020-00622-y
- https://www.nature.com/articles/s41598-022-09954-8