Application value of a deep learning model based on 3D ultrasound videos in assisting radiologists of different experience levels to differentiate benign and malignant breast masses

doi:10.16150/j.1671-2870.2025.03.010

Abstract

Abstract:

Objective To establish a deep learning model based on 3D ultrasound videos (DL-3DUV) and investigate its application value in assisting radiologists with different levels of experience to differentiate benign and malignant breast masses. Methods The ResNet50 model was employed to develop DL-3DUV for the classification of benign and malignant breast masses. A retrospective study was conducted using automated breast volume scanner (ABVS) dynamic videos from 400 patients with breast masses (a total of 525 lesions), which were randomly divided into training and testing sets at an 8∶2 ratio. The diagnostic performance of DL-3DUV was compared with that of senior and junior radiologists, both independently and in combination. Results When diagnosing independently, DL-3DUV demonstrated comparable sensitivi-ty (83.33% vs. 81.77%), accuracy (82.50% vs. 84.60%), and area under the curve (AUC) (0.83 vs. 0.85) compared to senior radiologists (all P>0.05), though its specificity was significantly lower (81.58% vs. 87.73%, P<0.05). Compared with junior radiologists, DL-3DUV showed significantly higher sensitivity (83.33% vs. 78.60%), specificity (81.58% vs. 57.00%), accuracy (82.50% vs. 68.37%), and AUC (0.83 vs. 0.68) (P<0.05). The combination of senior radiologists and DL-3DUV achieved higher sensitivity (89.70% vs. 81.77%) and AUC (0.91 vs. 0.85) than senior radiologists alone (all P<0.05), with no significant differences in specificity (91.23% vs. 87.73%) or accuracy (89.17% vs. 84.60%) (all P>0.05). Similarly, the combination of junior radiologists and DL-3DUV significantly improved diagnostic performance compared to junior radiologists alone, with statistically significant differences in sensitivity (88.10% vs. 78.60%), specificity (82.47% vs. 57.00%), accuracy (85.47% vs. 68.37%), and AUC (0.85 vs. 0.68) (all P<0.05). Conclusions DL-3DUV exhibits significant value in differentiating benign and malignant breast masses and is expected to become a useful tool to assist ultrasonographers, particularly for junior radiologists.

Key words: Automated breast volume scanner, Breast cancer, Deep learning

CLC Number:

GUO Yuqing, WANG Changyan, LIU Yinchun, PANG Yun, ZHU Xia, GE Rui, LI Weiping, ZHANG Qi, CHEN Lin. Application value of a deep learning model based on 3D ultrasound videos in assisting radiologists of different experience levels to differentiate benign and malignant breast masses[J]. Journal of Diagnostics Concepts & Practice, 2025, 24(03): 312-319.

Figures/Tables 7

Figure 1

Figure 2

Figure 3

Figure 4

Table 1

Figure 5

Figure 6

References 24

[1]	邬昊, 吕青. 全球及中国乳腺癌的流行病学趋势及防控启示:2018-2022年《全球癌症统计报告》解读[J]. 中国普外基础与临床杂志, 2024, 31(7):796-802.
	WU H, LÜ Q. Epidemiological trends and implications of breast cancer prevention and control in China and the word: an interpretation of the global cancer statistics report 2018-2022[J]. Chin J Bases Clin Gen Surg, 2024, 31(7):796-802.
[2]	刘迎春, 陈林, 庞芸, 等. 基于自动乳腺容积扫描、乳腺X线摄影及MRI的BI-RADS分类鉴别乳腺良恶性肿[J]. 中国医学影像技术, 2021, 37(3): 386-390.
	LIU Y C, CHEN L, PANG Y, et al. BI-RADS classification based on automated breast volume scanning, X-ray mammography and MRI for differentiation of benign and malignant breast masses[J]. Chin J Med Imag Technol, 2021, 37(3):386-390.
[3]	CHEN L, CHEN Y, DIAO X H, et al. Comparative study of automated breast 3D ultrasound and handheld B-mode ultrasound for differentiation of benign and malignant breast masses[J]. Ultrasound Med Biol, 2013, 39(10):1735-1742.
[4]	中国超声医学工程学会仪器工程开发专业委员会, 中国抗癌协会肿瘤影像专委会, 中华医学会超声医学分会浅表组织与血管学组. 自动乳腺容积超声技术专家共识(2022版)[J]. 中国超声医学杂志, 2022, 38(3):241-247.
	The Instrument Engineering Development Professional Committee of the Chinese Society of Ultrasonic Medical Engineering, the Tumor Imaging Special Committee of the Chinese Anti Cancer Association, the Superficial Tissue and Vascular Group of the Ultrasound Medicine Branch of the Chinese Medical Association. Expert consensus on automated breast ultrasound technology (2022 edition)[J]. Chin J Ultrasound Med, 2022, 38(3):241-247.
[5]	BREM RF, TABAR L, DUFFY SW, et al. Assessing improvement in detection of breast cancer with three-dimensional automated breast US in women with dense breast tissue: the SomoInsight study[J]. Radiology, 2015, 274(3):663-673. doi: 10.1148/radiol.14132832 pmid: 25329763
[6]	VAN ZELST J C M, TAN T, PLATEL B, et al. Improved cancer detection in automated breast ultrasound by radio-logists using computer aided detection[J]. Eur J Radiol, 2017,89:54-59.
[7]	WANG Q, CHEN H, LUO G, et al. Performance of novel deep learning network with the incorporation of the automatic segmentation network for diagnosis of breast cancer in automated breast ultrasound[J]. Eur Radiol, 2022, 32(10):7163-7172.
[8]	WANG Y, CHOI E J, CHOI Y, et al. Breast cancer classification in automated breast ultrasound using multi-view convolutional neural network with transfer learning[J]. Ultrasound Med Biol, 2020, 46(5):1119-1132.
[9]	XIANG H, HUANG Y S, LEE C H, et al. 3-D Res-CapsNet convolutional neural network on automated breast ultrasound tumor diagnosis[J]. Eur J Radiol, 2021,138:109608.
[10]	ZHUANG Z, DING W, ZHUANG S, et al. Tumor classification in automated breast ultrasound (ABUS) based on a modified extracting feature network[J]. Comput Med Ima-ging Graph, 2021,90:101925.
[11]	ZHOU Y, CHEN H, LI Y, et al. Multi-task learning for segmentation and classification of tumors in 3D automated breast ultrasound images[J]. Med Image Anal, 2021,70:101918.
[12]	RYAN J T, HAYGOOD T M, YAMAL J M, et al. The "memory effect" for repeated radiologic observations[J]. Am J Roentgenol, 2011, 197(6):W985-W991.
[13]	D'ORSI C, MORRIS E, MENDELSON E. ACR BI-RADS Atlas: Breast Imaging Reporting and Data System[M]. American Collegeof Radiology, 2013.
[14]	WANG A, WU Y, JI X, et al. Assessing and optimizing large language models on spondyloarthritis multi-choice question answering: protocol for enhancement and assessment[J]. JMIR Res Protoc, 2024,13:e57001.
[15]	XU X, BAO L, TAN Y, et al. 1000-case reader study of radiologists' performance in interpretation of automated breast volume scanner images with a computer-aided detection system[J]. Ultrasound Med Biol, 2018, 44(8):1694-1702. doi: S0301-5629(18)30184-4 pmid: 29853222
[16]	YANG S, GAO X, LIU L, et al. Performance and reading time of automated breast us with or without computer-aided detection[J]. Radiology, 2019, 292(3):540-549. doi: 10.1148/radiol.2019181816 pmid: 31210612
[17]	HE P, CUI L G, CHEN W, et al. Subcategorization of ultrasonographic BI-RADS category 4: assessment of diagnostic accuracy in diagnosing breast lesions and influence of clinical factors on positive predictive value[J]. Ultrasound Med Biol, 2019, 45(5):1253-1258. doi: S0301-5629(18)30555-6 pmid: 30799123
[18]	麦武平, 周美君, 胡叶, 等. AI超声辅助诊断系统在低年资医师诊断甲状腺结节良恶性中的应用[J]. 中国超声医学杂志, 2021, 37(3):248-251.
	MAI W P, ZHOU M J, HU Y, et al. The application of AI ultrasound-assisted diagnosis system in diagnosing benign and malignant thyroid nodules for junior sonographers[J]. Chin J Ultrasound Med, 2021, 37(3):248-251.
[19]	刘迎春, 陈林, 庞芸, 等. 自动乳腺容积扫描冠状面特征联合超声BI-RADS诊断乳腺癌的价值[J]. 中国临床医学影像杂志, 2020, 31(11):788-792.
	LIU Y C, CHEN L, PANG Y, et al. Values of combined coronal view features of automated breast scanning with ultrasound BI-RADS for the diagnosis of breast cancer[J]. J China Clin Med Imag, 2020, 31(11):788-792.
[24]	YANG J Y, SHEN J, SHEN S P, et al. Construction and analysis of preoperative lymph node metastasis load model of breast cancer patients based on dual-mode ultrasound[J]. Chin J Gen Pract, 2024, 22(4):646-650.
[25]	张杰秋, 伍棋, 王舰梅, 等. 基于机器学习的病理组学特征可预测乳腺癌患者对新辅助化疗的反应[J]. 重庆医科大学学报, 2023, 48(12):1483-1488.
	ZHANG J Q, WU Q, WANG J M, et al. Pathomics signature based on machine learning can predict the response to neoadjuvant chemotherapy in breast cancer patients[J]. J Chongqing Med Univ, 2023, 48(12):1483-1488.
[20]	包凌云, 朱罗茜, 孔凡雷, 等. 自动乳腺全容积成像和常规超声对乳腺微钙化诊断的对比研究[J]. 中华超声影像学杂志, 2012, 21(3):220-223.
	BAO L Y, ZHU L Q, KONG F L, et al. Comparative study of automated breast volume scanner with conventional ultrasound in diagnosing of breast microcalcifications[J]. Chin J Ultrasonography, 2012, 21(3):220-223.
[21]	VAN ZELST J C M, MANN R M. Automated three-dimensional breast US for screening: technique, artifacts, and lesion characterization[J]. Radiographics, 2018, 38(3):663-683. doi: 10.1148/rg.2018170162 pmid: 29624482
[22]	吴佩佩, 张刚, 李倩, 等. 基于超声构建乳腺癌腋下淋巴结良恶性的预测模型[J]. 安徽医学, 2024, 45(11):1417-1423.
	WU P P, ZHANG G, LI Q, et al. Prediction model for benign and malignant axillary lymph nodes in breast cancer based on ultrasound[J]. Anhui Med, 2024, 45(11): 1417-1423.
[23]	安林林, 马芳, 沈莉莉, 等. 乳腺癌超声征象与免疫组化指标表达关系的研究进展[J]. 中国临床研究, 2024, 37(10):1489-1493.
	AN L L, MA F, SHEN L L, et al. Research progress on the relationship between ultrasound signs and expression of immunohistechemical indicators in breast cancer[J]. Chin Clin Res, 2024, 37(10):1489-1493.
[24]	杨骏宇, 沈吉, 沈思平, 等. 基于双模态超声的乳腺癌患者术前淋巴结转移负荷模型构建分析[J]. 中华全科医学, 2024, 22(4):646-650.

Diagnostic method	Sensitivity (%)	Specificity (%)	Accurcy (%)	AUC
DL-3DUV	83.33(76.36, 95.30)	81.58(73.09, 90.07)	82.50(74.17, 90.83)	0.83(0.73, 0.92)
Senior radiologists	81.77(73.31, 90.23)	87.73(80.54, 94.92)	84.60(75.95, 93.25)	0.85(0.80, 0.90)
Junior radiologists	78.60(69.61, 87.59)	57.00(46.15, 67.85)	68.37(58.18, 78.56)	0.68(0.61, 0.75)
Senior radiologists+ DL-3DUV	89.70(83.04, 96.31)	91.23(85.03, 97.43)	89.17(82.36, 95.98)	0.91(0.87, 0.95)
Junior radiologists+ DL-3DUV	88.10(81.00, 95.20)	82.47(74.14, 90.80)	85.47(77.75, 93.19)	0.85(0.80, 0.91)
Comparison between radiologists and DL-3DUV
Senior radiologists vs. DL-3DUV（P）	0.500	0.016	0.400	0.071
Junior radiologists vs. DL-3DUV（P）	0.031	<0.001	<0.001	<0.001
Senior radiologists vs. junior radiologists（P）	0.125	<0.001	<0.001	<0.001
Comparison between radiologists in combination with DL-3DUV and radiologists
Senior radiologists+ DL-3DUV vs. senior radiologists（P）	0.001	0.125	0.118	0.001
Junior radiologists+ DL-3DUV vs. junior radiologists（P）	<0.001	<0.001	0.012	<0.001
Comparison between radiologists in combination with DL-3DUV and DL-3DUV
Senior radiologists+ DL-3DUV vs. DL-3DUV（P）	0.004	0.001	0.824	<0.001
Junior radiologists+ DL-3DUV vs. DL-3DUV（P）	0.031	>0.99	0.125	0.007
Comparison of different radiologists in combination with DL-3DUV
Senior radiologists+ DL-3DUV vs. junior radiologists+ DL- 3DUV（P）	0.250	0.002	0.092	0.001