A machine learning model for detecting invasive ductal carcinoma with Google Cloud AutoML Vision

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Accurate identification of invasive ductal carcinoma (IDC) is crucial for breast cancer diagnosis and treatment.
• Whole slide imaging (WSI) generates large datasets that require advanced analytical tools.
• Traditional machine learning model development can be resource-intensive.

Purpose of the Study:

• To evaluate the feasibility of using automated machine learning (AutoML) for IDC detection in WSI.
• To develop and test an ML model for automated IDC identification.
• To compare AutoML performance against existing methods.

Main Methods:

• An experimental machine learning (ML) model was developed using Google Cloud AutoML Vision.
• A large public dataset of 278,124 histopathology images was utilized.
• Data augmentation techniques, including image rotation, were applied to balance IDC sample distribution, resulting in 378,215 images.

Main Results:

• The AutoML model achieved an average accuracy of 91.6% (Area Under Precision-Recall Curve).
• A balanced accuracy of 84.6% was obtained on a held-out test dataset.
• The model demonstrated comparable performance on newly collected hospital breast tissue samples, indicating generalization capability.

Conclusions:

• AutoML technology is mature and feasible for the automated identification of IDC in WSI.
• Combining AutoML with cloud computing offers significant advantages for large-scale medical image analysis.
• The findings support the integration of AutoML into digital pathology workflows for improved diagnostic efficiency.