Satake Hiroko, Kinoshita Fumie, Ishigaki Satoko, Kato Keita, Jo Yusuke, Shimada Satoko, Masuda Norikazu, Naganawa Shinji
Department of Radiology, Nagoya University Graduate School of Medicine, Nagoya, Aichi, Japan.
Division of Data Science, Data Coordinating Center, Department of Advanced Medicine, Nagoya University Hospital, Nagoya, Aichi, Japan.
Magn Reson Med Sci. 2024 May 15. doi: 10.2463/mrms.mp.2023-0168.
To investigate the predictive performance of radiomic features extracted from breast MRI for upgrade of ductal carcinoma in situ (DCIS) to invasive carcinoma.
This retrospective study included 71 women with DCIS lesions diagnosed preoperatively by biopsy. All women underwent breast dynamic contrast-enhanced (DCE) MRI of the breast, which included pre-contrast and five post-contrast phases continuously with a time resolution of 60s. Lesion segmentation was performed manually, and 144 radiomic features of the lesions were extracted from T2-weighted images (T2WI), pre-contrast T1-weighted images (T1WI), and post-contrast 1st, 2nd, and 5th phase subtraction images on DCE-MRI. Qualitative features of mammography, ultrasound, and MRI were also assessed. Clinicopathological features were evaluated using medical records. The least absolute shrinkage and selection operator (LASSO) algorithm was applied for features selection and model building. The predictive performance of postoperative upgrade to invasive carcinoma was assessed using the area under the receiver operating characteristic curve.
Surgical specimens revealed 13 lesions (18.3%) that were upgraded to invasive carcinoma. Among clinicopathological and qualitative features, age was the only significant predictive variable. No significant radiomic features were observed on T2WI and post-contrast 2nd phase subtraction images on DCE-MRI. The area under the curves (AUCs) of the LASSO radiomics model integrated with age were 0.915 for pre-contrast T1WI, 0.862 for post-contrast 1st phase subtraction images, and 0.833 for post-contrast 5th phase subtraction images. The AUCs of the 200-times bootstrap internal validations were 0.885, 0.832, and 0.775.
A radiomics approach using breast MRI may be a promising method for predicting the postoperative upgrade of DCIS. The present study showed that the radiomic features extracted from pre-contrast T1WI and post-contrast subtraction images in the very early phase of DCE-MRI were more predictable.
研究从乳腺MRI中提取的放射组学特征对导管原位癌（DCIS）进展为浸润性癌的预测性能。
这项回顾性研究纳入了71例术前经活检诊断为DCIS病变的女性。所有女性均接受了乳腺动态对比增强（DCE）MRI检查，包括平扫及5期连续增强扫描，时间分辨率为60秒。手动进行病变分割，并从T2加权图像（T2WI）、平扫T1加权图像（T1WI）以及DCE-MRI的增强后第1、2和5期减影图像中提取病变的144个放射组学特征。还评估了乳腺X线摄影、超声和MRI的定性特征。使用病历评估临床病理特征。应用最小绝对收缩和选择算子（LASSO）算法进行特征选择和模型构建。使用受试者操作特征曲线下面积评估术后进展为浸润性癌的预测性能。
手术标本显示13个病变（18.3%）进展为浸润性癌。在临床病理和定性特征中，年龄是唯一显著的预测变量。在T2WI和DCE-MRI增强后第2期减影图像上未观察到显著的放射组学特征。结合年龄的LASSO放射组学模型在平扫T1WI上的曲线下面积（AUC）为0.915，增强后第1期减影图像为0.862，增强后第5期减影图像为0.833。200次自举内部验证的AUC分别为0.885、0.832和0.775。
使用乳腺MRI的放射组学方法可能是预测DCIS术后进展的一种有前景的方法。本研究表明，从DCE-MRI早期平扫T1WI和增强后减影图像中提取的放射组学特征具有更高的预测性。
