State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine, Sun Yat-sen University Cancer Center, Guangzhou, China.
Department of Ultrasound, Sun Yat-Sen University Cancer Center, Guangzhou, China.
Br J Radiol. 2022 Feb 1;95(1130):20210438. doi: 10.1259/bjr.20210438. Epub 2021 Dec 15.
The aim of this study was to investigate the detection efficacy of deep learning (DL) for automatic breast ultrasound (ABUS) and factors affecting its efficacy.
Females who underwent ABUS and handheld ultrasound from May 2016 to June 2017 ( = 397) were enrolled and divided into training ( = 163 patients with breast cancer and 33 with benign lesions), test ( = 57) and control ( = 144) groups. A convolutional neural network was optimized to detect lesions in ABUS. The sensitivity and false positives (FPs) were evaluated and compared for different breast tissue compositions, lesion sizes, morphologies and echo patterns.
In the training set, with 688 lesion regions (LRs), the network achieved sensitivities of 93.8%, 97.2% and 100%, based on volume, lesion and patient, respectively, with 1.9 FPs per volume. In the test group with 247 LRs, the sensitivities were 92.7%, 94.5% and 96.5%, respectively, with 2.4 FPs per volume. The control group, with 900 volumes, showed 0.24 FPs per volume. The sensitivity was 98% for lesions > 1 cm, but 87% for those ≤1 cm ( < 0.05). Similar sensitivities and FPs were observed for different breast tissue compositions (homogeneous, 97.5%, 2.1; heterogeneous, 93.6%, 2.1), lesion morphologies (mass, 96.3%, 2.1; non-mass, 95.8%, 2.0) and echo patterns (homogeneous, 96.1%, 2.1; heterogeneous 96.8%, 2.1).
DL had high detection sensitivity with a low FP but was affected by lesion size.
DL is technically feasible for the automatic detection of lesions in ABUS.
本研究旨在探讨深度学习（DL）在自动乳腺超声（ABUS）中的检测效能及其影响因素。
纳入 2016 年 5 月至 2017 年 6 月期间接受 ABUS 和手持超声检查的女性（=397 例），并将其分为训练组（=163 例乳腺癌患者和 33 例良性病变患者）、测试组（=57 例）和对照组（=144 例）。优化卷积神经网络以检测 ABUS 中的病变。评估并比较不同乳腺组织成分、病变大小、形态和回声模式下的敏感性和假阳性（FP）。
在训练组中，该网络对 688 个病变区域（LR）的检测灵敏度分别为基于体积、LR 和患者的 93.8%、97.2%和 100%，每个体积的 FP 为 1.9 个。在 247 个 LR 的测试组中，其灵敏度分别为 92.7%、94.5%和 96.5%，每个体积的 FP 为 2.4 个。900 个体积的对照组的每个体积的 FP 为 0.24 个。对于>1cm 的病变，其灵敏度为 98%，而对于≤1cm 的病变，其灵敏度为 87%（<0.05）。对于不同的乳腺组织成分（均匀性：97.5%，2.1；异质性：93.6%，2.1）、病变形态（肿块：96.3%，2.1；非肿块：95.8%，2.0）和回声模式（均匀性：96.1%，2.1；异质性：96.8%，2.1），其灵敏度和 FP 均相似。
DL 具有高检测灵敏度和低 FP，但受病变大小的影响。
DL 技术在 ABUS 中自动检测病变是可行的。
