Sharpness-aware minimization improves bacterial classification accuracy
Sharpness-Aware Minimization (SAM) Improves Classification Accuracy of Bacterial Raman Spectral Data Enabling Portable Diagnostics
Machine Learning
Summary
When bacteria become resistant to antibiotics, it can cause serious health problems, especially where medical resources are limited. Raman spectroscopy is a quick way to identify bacteria and their resistance but is hard to analyze accurately because the data can be noisy and limited. The authors showed that using a training method called Sharpness-Aware Minimization (SAM) helps computer models better recognize bacteria from this data by improving accuracy. This could help build portable devices that diagnose bacterial infections more reliably and faster.
What this means in practice
- •For clinical diagnostics teams: Improve bacterial species identification accuracy during antibiotic resistance testing using Raman spectral data and SAM-optimized models.
- •For portable medical device engineers: Develop smaller, faster, and more reliable bacterial diagnostic tools by leveraging SAM to handle noisy Raman spectra with less preprocessing.
Authors
Kaitlin Zareno, Jarett Dewbury, Siamak K. Sorooshyari, Hossein Mobahi, Loza F. Tadesse
Abstract
Antimicrobial resistance is expected to claim 10 million lives per year by 2050, and resource-limited regions are most affected. Raman spectroscopy is a novel pathogen diagnostic approach promising rapid and portable antibiotic resistance testing within a few hours, compared to days when using gold standard methods. However, current algorithms for Raman spectra analysis 1) are unable to generalize well on limited datasets across diverse patient populations and 2) require increased complexity due to the necessity of non-trivial pre-processing steps, such as feature extraction, which are essential to mitigate the low-quality nature of Raman spectral data. In this work, we address these limitations using Sharpness-Aware Minimization (SAM) to enhance model generalization across a diverse array of hyperparameters in clinical bacterial isolate classification tasks. We demonstrate that SAM achieves accuracy improvements of up to 10.5% on a single split, and an increase in average accuracy of 2.7% across all splits in spectral classification tasks over the traditional optimizer, Adam. These results display the capability of SAM to advance the clinical application of AI-powered Raman spectroscopy tools.