Computer Vision For Bacterial Gram Stain Classification
CWSF · 2026 Disease & Illness Gold Medal
Overview
Antibiotic resistance is one of the most urgent crises in global health, and the first step in fighting it is correctly identifying whether a bacterium is Gram-positive or Gram-negative — a distinction that directly determines which antibiotics will work. Manual microscopy interpretation is slow, subjective, and inaccessible in resource-limited settings. This project developed four progressively optimized convolutional neural networks (CNNs) trained on 8,000 labeled Gram-stained microscopy images. The final model hit 98.33% test accuracy, and I reproduced that result across two independent runs. These results put us ahead of the 94.9% benchmark reported by Smith et al. (Harvard Medical School, 2018) and the 95.1% benchmark reported by Hee Kim et al.(Heidelberg University, 2023). Grad-CAM visualization confirms the model learns genuine bacterial features, rather than background or other artifacts in the slide. This demonstrates that reliable, automated Gram stain classification is achievable without specialized equipment or proprietary data.
Awards (2)
- Gold Medal
- Selected for CWSF 2026
Competition history
- CWSF 2026
Related projects
CWSF · 2026
DEEP-GRAM: A Deep Learning Model for Gram Stain Species Prediction in Bloodstream Infections
ISEF · 2022
Automatic Classification of Peripheral Neutrophils on Digital Images Analyzed by Artificial Intelligence
ISEF · 2023
MicroScan: A Computer Vision Tool To Assist Malaria Microscopy Diagnosis
ISEF · 2025
Convolutional Neural Networks Applied to Hematological Analysis
Closest projects by meaning, across every fair and year in the corpus.