Deep learning for image classification in dedicated breast positron emission tomography (dbPET).

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Dedicated breast positron emission tomography (dbPET) is an emerging imaging modality for breast cancer (BC) detection.
• Accurate BC diagnosis relies on expert interpretation of complex imaging data.
• Deep learning (DL) offers potential for automating and enhancing diagnostic accuracy in medical imaging.

Purpose of the Study:

• To identify the optimal deep learning (DL) model for breast cancer (BC) prediction using dedicated breast positron emission tomography (dbPET) images.
• To compare the diagnostic performance of the developed DL model against human expert interpretation.

Main Methods:

• A retrospective analysis of 618 dbPET examinations from 284 women (diagnosed with BC or non-BC).
• Training a Xception-based DL model on dbPET images (maximum intensity projections) from 458 breasts.
• Evaluating the DL model's sensitivity, specificity, and AUC against two expert radiologists and two radiology residents.

Main Results:

• The DL model achieved 93% sensitivity and 93% specificity.
• The DL model's AUC (0.937) was comparable to expert radiologists (AUCs 0.983 and 0.941).
• The DL model's performance showed a trend towards superiority over radiology residents (AUCs 0.876 and 0.868).

Conclusions:

• The developed deep learning model demonstrates significant potential for breast cancer detection using dbPET imaging.
• The DL model achieves diagnostic performance on par with experienced radiologists.
• This AI tool could serve as a valuable aid in clinical practice for dbPET image interpretation.