A Pilot Study of Breast Cancer Histopathological Image Classification Using Google Teachable Machine: A No-Code
Namit Singla1, Abhra Ghosh2, Manthan Dhingra3
1Medicine, Dayanand Medical College and Hospital, Ludhiana, IND.
Cureus
|August 5, 2025
Summary
This pilot study shows Google Teachable Machine (GTM) can classify breast histopathology images with preliminary feasibility. While accurate for normal and in situ categories, further development is needed for invasive carcinoma detection.
Area of Science:
- Digital Pathology
- Artificial Intelligence in Medicine
- Computational Pathology
Background:
- Breast cancer histopathology is crucial for diagnosis but faces inter-observer variability and workload challenges.
- Innovative solutions are needed to improve diagnostic accuracy and efficiency in breast cancer pathology.
- Artificial intelligence (AI) offers potential for automating image analysis and aiding pathologists.
Purpose of the Study:
- To assess the feasibility and diagnostic performance of a no-code AI platform, Google Teachable Machine (GTM), for classifying breast histopathology images.
- To evaluate GTM's ability to categorize images into normal, benign, in situ carcinoma, and invasive carcinoma.
- To determine the accuracy, precision, recall, and F1-score of the GTM model on both internal and external validation datasets.
Main Methods:
- Utilized 380 hematoxylin and eosin-stained breast histopathology images from an open-access repository.
- Trained a Google Teachable Machine (GTM) model using 85% of the data (50 epochs, batch size 16, learning rate 0.0001).
- Externally validated the trained GTM model on 39 independent images, calculating standard performance metrics.
Main Results:
- The GTM model achieved 88.3% internal validation accuracy, with high per-class accuracies (87-93%).
- External validation on 39 images yielded an overall accuracy of 76.9% and a macro-averaged F1-score of 0.77.
- The model showed high precision but reduced recall for invasive carcinoma, indicating challenges in distinguishing invasive from non-invasive lesions with limited data.
Conclusions:
- Google Teachable Machine (GTM) demonstrates preliminary feasibility for multi-class breast histopathology classification without coding expertise.
- The AI platform shows promise, particularly for normal and in situ carcinoma categories, but requires further refinement.
- Enhancing clinical applicability necessitates larger datasets, advanced AI architectures, and explainable AI methods to improve sensitivity for invasive carcinoma.
