Virtual Biopsy by Using Artificial Intelligence-based Multimodal Modeling of Binational Mammography Data.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Computational models using artificial intelligence (AI) are increasingly utilized for diagnosing malignant breast lesions.
• Accurately assessing specific pathologic lesion subtypes from radiologic images, as confirmed by biopsy, remains a significant challenge.

Purpose of the Study:

• To develop and validate an AI-based model for identifying breast lesion subtypes.
• The model integrates mammographic images with electronic health records containing histopathologic diagnoses.

Main Methods:

• A retrospective study involving 26,569 mammograms from 9,234 women was used for algorithm pretraining.
• A hybrid model combining convolutional neural networks (CNNs) and supervised learning was trained on data from women in Israel and the United States.
• Performance was evaluated using the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals (CIs).

Main Results:

• The AI model achieved an AUC of 0.88 for the Israeli test set and 0.80 for the U.S. test set in predicting malignancy (ductal carcinoma in situ or invasive cancer).
• The Israeli model was validated in 456 women and tested in 441; the U.S. model was validated in 350 and tested in 344.
• Significant differences in AUC between the Israeli and U.S. models were observed (P = .006), highlighting the need for further generalizability studies.

Conclusions:

• AI applied to large datasets of clinical and mammographic images shows promise in identifying breast lesion subtypes.
• This technology may enhance diagnostic workflows and help minimize errors associated with biopsy sampling.
• Further investigation is required to confirm the generalizability of these AI models across diverse patient populations.