Computer-aided detection of mammographic microcalcifications: pattern recognition with an artificial neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Computer-Aided Diagnosis

Background:

• Automated detection of clustered microcalcifications on mammograms is crucial for early breast cancer diagnosis.
• Existing detection programs often struggle with accuracy, leading to false positives and missed diagnoses.
• Convolutional Neural Networks (CNNs) show promise for image analysis tasks.

Purpose of the Study:

• To investigate the effectiveness of a CNN-based signal classifier for improving automated microcalcification detection accuracy.
• To evaluate the reduction in false positives achieved by integrating a CNN into an existing detection program.
• To assess the CNN's performance across different microcalcification visibility levels (obvious, average, subtle).

Main Methods:

• Fifty-two mammograms with clustered microcalcifications were analyzed.
• Mammograms were categorized into obvious, average, and subtle groups; average and subtle groups were combined for training/testing.
• Regions of Interest (ROIs) were identified, and CNNs were trained using a back-propagation method. Classification accuracy was assessed using ROC analysis.

Main Results:

• CNNs achieved a high classification performance with an area under the ROC curve (Az) of 0.9.
• Incorporating the CNN classifier reduced false-positive clusters by over 70% across various true-positive rates.
• For obvious cases, the false-positive rate decreased from 0.35 to 0.1 per image at 100% true-positive rate. Average/subtle cases saw improved detection accuracy.

Conclusions:

• CNN-based signal classification significantly enhances the accuracy of automated microcalcification detection programs.
• The developed CNN classifier effectively reduces false positives, potentially improving radiologist workflow and patient outcomes.
• This approach demonstrates a viable method for improving computer-aided detection of subtle and average clustered microcalcifications.