Tao Xuetong, Gandomkar Ziba, Li Tong, Brennan Patrick C, Reed Warren M
Discipline of Medical Imaging Science, Faculty of Health Sciences, Western Ave, Camperdown NSW 2050, Australia.
The Daffodil Centre, The University of Sydney, A Joint Venture with Cancer Council NSW, Sydney, NSW 2006, Australia.
Br J Radiol. 2025 Jan 1;98(1165):75-88. doi: 10.1093/bjr/tqae195.
This study aims to investigate radiologists' interpretation errors when reading dense screening mammograms using a radiomics-based artificial intelligence approach.
Thirty-six radiologists from China and Australia read 60 dense mammograms. For each cohort, we identified normal areas that looked suspicious of cancer and the malignant areas containing cancers. Then radiomic features were extracted from these identified areas and random forest models were trained to recognize the areas that were most frequently linked to diagnostic errors within each cohort. The performance of the model and discriminatory power of significant radiomic features were assessed.
We found that in the Chinese cohort, the AUC values for predicting false positives were 0.864 (CC) and 0.829 (MLO), while in the Australian cohort, they were 0.652 (CC) and 0.747 (MLO). For false negatives, the AUC values in the Chinese cohort were 0.677 (CC) and 0.673 (MLO), and in the Australian cohort, they were 0.600 (CC) and 0.505 (MLO). In both cohorts, regions with higher Gabor and maximum response filter outputs were more prone to false positives, while areas with significant intensity changes and coarse textures were more likely to yield false negatives.
This cohort-based pipeline proves effective in identifying common errors for specific reader cohorts based on image-derived radiomic features.
This study demonstrates that radiomics-based AI can effectively identify and predict radiologists' interpretation errors in dense mammograms, with distinct radiomic features linked to false positives and false negatives in Chinese and Australian cohorts.
本研究旨在使用基于放射组学的人工智能方法，调查放射科医生在阅读致密型乳腺筛查钼靶片时的解读错误。
来自中国和澳大利亚的36名放射科医生阅读了60张致密型乳腺钼靶片。对于每个队列，我们确定了看似可疑癌症的正常区域和包含癌症的恶性区域。然后从这些确定的区域中提取放射组学特征，并训练随机森林模型以识别每个队列中最常与诊断错误相关的区域。评估了模型的性能和显著放射组学特征的鉴别能力。
我们发现，在中国队列中，预测假阳性的AUC值在CC位为0.864，在MLO位为0.829；而在澳大利亚队列中，CC位为0.652，MLO位为0.747。对于假阴性，中国队列中的AUC值在CC位为0.677，MLO位为0.673；澳大利亚队列中，CC位为0.600，MLO位为0.505。在两个队列中，Gabor和最大响应滤波器输出较高的区域更容易出现假阳性，而强度变化显著和纹理粗糙的区域更有可能产生假阴性。
这种基于队列的流程被证明可有效地根据图像衍生的放射组学特征识别特定读者队列的常见错误。
本研究表明，基于放射组学的人工智能可以有效地识别和预测放射科医生在致密型乳腺钼靶片中的解读错误，在中国和澳大利亚队列中，不同的放射组学特征与假阳性和假阴性相关。
