Dynamic AI-assisted ipsilateral tissue matching for digital breast tomosynthesis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Breast Cancer Detection

Background:

• Accurate lesion localization is crucial for effective breast cancer diagnosis and treatment.
• Digital breast tomosynthesis (DBT) has improved breast cancer detection rates.
• Localization errors can occur, especially for non-expert readers, potentially leading to missed lesions.

Purpose of the Study:

• To evaluate the effectiveness of AI-assisted ipsilateral tissue matching in DBT for reducing localization errors.
• To assess the impact of AI on localization accuracy, particularly for non-expert radiologists.
• To determine if AI assistance can minimize localization errors beyond typical tumor boundaries.

Main Methods:

• Two-part study involving radiologists evaluating an AI tool for digital breast tomosynthesis (DBT).
• Part 1: Subjective evaluation of AI confidence and usefulness in 11 cases.
• Part 2: Lesion annotation with and without AI assistance in 30 cases, measuring localization errors (RMSE, MDE) against expert consensus.

Main Results:

• Radiologists reported increased confidence and usefulness (6.21/10) with AI assistance (p < 0.001).
• Localization errors (RMSE, MDE) were significantly higher without AI for abnormal lesions (p < 0.05).
• Non-expert readers showed >60% reduction in RMSE and MDE with AI, bringing errors within clinically relevant tumor dimensions.

Conclusions:

• AI-assisted tissue matching in DBT significantly enhances localization accuracy.
• The AI tool provides particular benefit to non-expert radiologists and in complex cases.
• AI assistance reduces localization errors to within typical tumor sizes, potentially improving lesion detection and preventing missed diagnoses.