By SALMA ED-DAOUI
Introduction
Italian researchers from the Italian Institute of Technology in Genoa have made significant strides in the field of breast cancer prediction with the use of an artificial intelligence (AI) model. Their study, published in Radiology on June 13, reveals promising results that surpass traditional measures with a new AI model for breast cancer prediction after negative screening results.
Higher Performance than Breast Density Measures
The team, led by Celeste Damiani, Ph.D., investigated the performance of the AI model in comparison to breast density measures. They discovered that the AI model demonstrated higher performance in predicting breast cancer risk compared to relying solely on breast density information.
Performance for Different Types of Cancer
Furthermore, the researchers found that the AI model performed well in predicting both invasive cancer and ductal carcinoma in situ (DCIS). In particular, it showed even higher performance in predicting advanced cancer risk and screen-detected estrogen receptor-positive cancer.
Role in Risk-Based Screening Algorithms
Based on their findings, Damiani and her colleagues advise that the AI model may have a considerable role in the development of risk-based screening algorithms. These algorithms can enhance the truth and strength of breast cancer screening programs.
The Potential of Additional Mammographic Information
While mammographic breast density is a known predictor of future breast cancer risk, the researchers hypothesized that incorporating additional mammographic information, such as computer-aided detection (CAD) suspicion scores, could improve the assessment of breast cancer risk.
Building on Previous AI Model Success
The AI model used in the study, called Mirai, was developed by researchers at the Massachusetts Institute of Technology and has previously shown success in predicting breast cancer risk over five years.
Study Design and Data Analysis
To evaluate the AI model’s performance, the team utilized data from the U.K.’s National Health Service (NHS) Breast Screening Program. The dataset included women with screen-detected cancer, both invasive and DCIS, as well as interval breast cancer cases.
Promising Results: AI Model Outperforms Breast Density Measures
The results of the study demonstrated that the AI model had a significantly higher likelihood of predicting hereafter front malignant neoplastic disease compared to models that relied solely on front-density data. The model achieved an area below the curve (AUC) of 0.67, outperforming breast denseness measures with an AUC of 0.56 (p < 0.001).
Improvement in Interval Cancer Prediction
While the addition of breast density improved the model’s performance in predicting interval cancers (those detected between screening appointments) with an AUC increase from 0.69 to 0.71 (p < 0.001), no significant improvement was observed for screen-detected cancers.
Similar Performance for Different Cancer Types
The AI model exhibited similar performances for detecting both invasive cancer and DCIS, with AUC values of 0.68 and 0.66, respectively (p = 0.057).
Further Research and Future Directions
Although the study yielded promising results, the authors emphasize the importance of further research to explain the internal workings of the AI model and determine the specific mammographic features contributing to its performance. They also stress the need for additional testing in retrospective and prospective studies to compare the model with other tools, particularly those involving digital breast tomosynthesis (DBT).
Editorial Perspective: Image-Based Risk Modeling and Optimal Screening Pathways
In an accompanying editorial, Ritse Mann, MD, PhD, and Ioannis Sechopoulos, Ph.D., from the Netherlands, highlight the potential of image-based risk modeling in determining optimal screening pathways. They suggest that the AI model’s ability to detect early signs of breast cancer may explain its superior performance compared to traditional risk estimators.
Conclusion:
The Italian team’s research demonstrates the potential of AI models in accurately predicting breast cancer risk. By surpassing the limitations of breast density measures, AI-based models can heighten risk judgment and meliorate the effectiveness of breast cancer screening programs. Continued research, transparency, and collaboration wish to be requisite in harnessing the full potentiality of AI technology for early detection, diagnosis, and personalized treatment of breast cancer
the study can be found here.