MedElevator Войти в исследования
← Поиск публикаций

DOI 10.1007/s00330-026-12536-1

Медицинская библиотека · European radiology · 2026

From pixels to pathology: how artificial intelligence mammographic risk scores capture tumor biology through imaging.

Zhang Zi, Elabeid Jafer, Ali Thowaiba, Pantleo Jennifer, Gonzalez Nelda, Parghi Chirag

СтатьяПеревод в очереди

Аннотация

OBJECTIVES

To characterize clinical-pathologic tumor features associated with artificial intelligence (AI)-generated risk scores from prior-year screening mammograms.

MATERIALS AND

METHODS

This retrospective study included women who underwent breast biopsy following a screening mammogram in 2021 across four U.S. states. AI risk scores were obtained from prior-year screening mammograms using an FDA-approved AI model. Receiver operating characteristic (ROC) analysis was used to evaluate the discriminative ability of AI risk scores. Among patients with breast cancer, linear regression was used to assess associations between prior-year risk scores and cancer characteristics.

RESULTS

Among 1509 patients included (mean age 58.56 ± 12.28 years), 508 (33.7%) had biopsy-confirmed breast cancer. The area under the ROC curve (AUC) for prior-year AI risk score predicting biopsy-confirmed cancer was 0.62 (95% CI: 0.59-0.65). In univariate analysis of biopsy-positive patients, invasive lobular carcinoma (ILC) had significantly higher prior-year AI risk scores than ductal carcinoma in situ (p = 0.009), while Grade 3 tumors had significantly lower scores than Grade 1 (p = 0.016). After adjusting for tumor grade, the ILC association was no longer significant (p = 0.136), suggesting that tumor grade may mediate this relationship; Grade 3 tumors showed a marginal association with lower risk scores (p = 0.068).

CONCLUSION

AI-generated risk scores from prior-year screening mammograms demonstrated modest discrimination between biopsy-confirmed malignant and non-malignant cases and may capture imaging features associated with low-grade tumors. Our findings suggest AI risk scores may reflect subtle imaging patterns of low-grade malignancy 1 year before clinical detection, thereby enhancing our understanding of AI model behavior and informing future research on clinical utility.

KEY POINTS: Question What clinical-pathologic tumor characteristics are reflected in artificial intelligence (AI)-generated risk scores from prior-year screening mammograms among patients with biopsy-confirmed outcomes? Findings Prior-year AI risk scores demonstrated modest discrimination between biopsy-confirmed malignant and non-malignant cases and may capture imaging features associated with low-grade tumors. Clinical relevance AI-generated risk scores from prior-year screening mammograms may capture subtle imaging features of evolving low-grade malignancy 1 year before clinical detection, offering insights into AI model behavior, enhancing explainability for clinicians, and informing future research on clinical utility.

Для цитирования: Zhang Zi, Elabeid Jafer, Ali Thowaiba, Pantleo Jennifer, Gonzalez Nelda, Parghi Chirag. From pixels to pathology: how artificial intelligence mammographic risk scores capture tumor biology through imaging.. European radiology. 2026. DOI: 10.1007/s00330-026-12536-1.

Предметные рубрики MeSH

Подготовка и перевод публикации
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…

История · было → стало

Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.

Тип публикации указан по данным источника. Уровень доказательности автоматически не присваивается.