AI-Based Self-Supervised Learning Model Using Non-Contrast Breast MRI for Early Screening and Clinical Utility Evaluation

Официальное название

Construction of an Early Breast Cancer Screening Warning Model Based on Self-supervised Learning With Plain MRI Scans and Prospective Clinical Utility Evaluation

Обзор

Breast cancer is the most common malignant disease among women worldwide, with rising incidence and younger age at onset in China. Early detection is critical for improving survival, yet current screening methods such as mammography and ultrasound show limited sensitivity in Chinese women, particularly those with dense breast tissue. Contrast-enhanced MRI offers higher diagnostic performance but its use is limited by high costs, safety concerns with gadolinium-based contrast agents, and limited accessibility. This investigator-initiated trial aims to evaluate the clinical application of non-contrast multiparametric MRI, combined with advanced artificial intelligence algorithms, for the early detection and diagnosis of breast cancer. The study will collect MRI imaging data from multiple centers and integrate radiomic features across T2-weighted imaging, diffusion-weighted imaging, and apparent diffusion coefficient maps. A deep learning-based model will be developed and validated to improve lesion detection, differential diagnosis, and risk stratification. The ultimate goal of this project is to establish a safe, accurate, and scalable breast cancer screening pathway suitable for Chinese women. By reducing dependence on invasive procedures and contrast agents, and by leveraging AI for standardization and efficiency, this approach may significantly improve early detection rates and contribute to better patient outcomes.

Подробное описание

This is a prospective, investigator-initiated clinical study designed to evaluate the role of radiomics and artificial intelligence in non-invasive, early detection and diagnosis of breast cancer. While mammography and ultrasound are widely used as first-line screening methods, their sensitivity and specificity remain suboptimal in Chinese women, particularly in individuals with dense breast tissue. Contrast-enhanced MRI has demonstrated superior diagnostic performance, but its clinical utility is limited due to high costs, safety concerns related to gadolinium deposition, and limited availability in population-based screening programs.

To address these challenges, this study will focus on non-contrast multiparametric breast MRI, including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) mapping. Imaging data will be prospectively collected from multiple clinical sites. A radiomics pipeline will be established to extract high-dimensional features characterizing lesion morphology, texture, and diffusion properties. Furthermore, an artificial intelligence-based model, developed using deep learning and self-supervised learning frameworks, will be trained and validated for lesion detection, classification, and risk prediction.

The primary aim of this trial is to construct and validate an imaging biomarker for early breast cancer detection based on non-contrast MRI and AI. Secondary objectives include evaluation of diagnostic accuracy compared with conventional imaging modalities, analysis of model performance across different molecular subtypes of breast cancer, and exploration of its potential application in predicting treatment response and clinical outcomes.

The expected outcome of this study is to provide robust evidence supporting the clinical feasibility of AI-guided non-contrast MRI as a safe, cost-effective, and scalable tool for early breast cancer screening in Chinese women. This work has the potential to optimize screening strategies, reduce unnecessary invasive procedures, and ultimately improve patient prognosis.

Вмешательства

• Диагностический тест Non-contrast multiparametric breast MRI with AI-based radiomics analysis
  Participants will receive standardized non-contrast multiparametric breast MRI scans (T2WI, DWI, ADC). Imaging features will be extracted and analyzed using artificial intelligence-based radiomics and deep learning algorithms to improve early detection and diagnosis of breast cancer.
• Диагностический тест Standard radiologist reading of non-contrast multiparametric breast MRI
  Imaging data interpreted by trained radiologists following routine clinical practice, without AI assistance.

Первичные конечные точки

• Diagnostic accuracy of AI-based non-contrast multiparametric MRI for breast cancer detection [Срок оценки: Within 12 months of study enrollment]

Вторичные конечные точки (1)

• Sensitivity and specificity stratified by breast cancer molecular subtype [Срок оценки: Within 12 months of enrollment]

Критерии участия

• Inclusion Criteria:

• Female, age 30-70 years
• Completed breast MRI scan, including at least T2WI, DWI, and ADC sequences
• Multimodal data acquired within the same time window (≤90 days)
• A clear clinical outcome: pathologically confirmed or ≥12-24 months of negative follow-up
• The time window between imaging examination and outcome determination was ≤90 days
• Signed informed consent
• Exclusion Criteria:

• Absolute contraindications to MRI (pacemaker, cochlear implant, ocular metal foreign body, etc.)
• Pregnant or lactating women
• Recent history of breast surgery/radiotherapy (≤6 months) or imaging after neoadjuvant therapy
• Substandard image quality (severe motion artifact, signal-to-noise ratio below threshold)
• Incomplete clinical data or time window exceeded
• Known breast cancer metastasis or recurrence

Критерии приведены из реестра в оригинале (на английском). Окончательную оценку соответствия проводит исследовательский центр.

Здоровые добровольцы: Нет

Дизайн исследования

Распределение

Не применимо

Модель

Параллельные группы

Маскирование

Открытое

Основная цель

Диагностика

Центры проведения

Список центров уточняется — проверьте первичный протокол.

Идентификаторы

NCT: NCT07205276 · SAHZhejiangU-20250916 · 2025-0736

Первоисточники (государственные реестры)

• ClinicalTrials.gov (NIH, США)

Открыть это исследование на ClinicalTrials.gov ↗