Devising a deep neural network based mammography phantom image filtering algorithm using images obtained under mAs...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology Quality Assurance

Background:

• Mammography phantom image analysis is crucial for quality control.
• Manual interpretation of these images can be time-consuming and labor-intensive.
• Automating this process could improve efficiency and consistency.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) algorithm for filtering mammography phantom images.
• To assess the performance of DNN-based models in classifying phantom images as pass or fail.
• To evaluate the potential of these algorithms in reducing human workload.

Main Methods:

• VGG16-based multi-class and binary-class classifiers were trained on 543 generated mammography phantom images.
• Filtering algorithms were designed using the trained scoring models.
• External validation was performed on 61 phantom images from two medical institutions.

Main Results:

• Binary-class classifiers achieved an F1-score of 0.93 and an area under the receiver operating characteristic curve of 0.97.
• Multi-class classifiers achieved an F1-score of 0.69.
• Filtering algorithms successfully processed 69% (42 of 61) of external validation images without human intervention.

Conclusions:

• Deep neural network algorithms show significant potential for automating mammography phantom image interpretation.
• This AI-driven approach can effectively reduce the workload associated with quality control in mammography.
• Further development could lead to widespread adoption for improved efficiency and accuracy.