A Comparative Study of Multiple Deep Learning Models Based on Multi-Input Resolution for Breast Ultrasound Images

Wu, Huaiyu1; Ye, Xiuqin1; Jiang, Yitao2,3; Tian, Hongtian1; Yang, Keen1; Cui, Chen2,3; Shi, Siyuan2,3; Liu, Yan4,5,6; Huang, Sijing1; Chen, Jing1; Xu, Jinfeng1; Dong, Fajin1

2022-07-07

10.3389/fonc.2022.869421

Frontiers in Oncology   影响因子和分区

2234-943X

PurposeThe purpose of this study was to explore the performance of different parameter combinations of deep learning (DL) models (Xception, DenseNet121, MobileNet, ResNet50 and EfficientNetB0) and input image resolutions (REZs) (224 x 224, 320 x 320 and 488 x 488 pixels) for breast cancer diagnosis. MethodsThis multicenter study retrospectively studied gray-scale ultrasound breast images enrolled from two Chinese hospitals. The data are divided into training, validation, internal testing and external testing set. Three-hundreds images were randomly selected for the physician-AI comparison. The Wilcoxon test was used to compare the diagnose error of physicians and models under P=0.05 and 0.10 significance level. The specificity, sensitivity, accuracy, area under the curve (AUC) were used as primary evaluation metrics. ResultsA total of 13,684 images of 3447 female patients are finally included. In external test the 224 and 320 REZ achieve the best performance in MobileNet and EfficientNetB0 respectively (AUC: 0.893 and 0.907). Meanwhile, 448 REZ achieve the best performance in Xception, DenseNet121 and ResNet50 (AUC: 0.900, 0.883 and 0.871 respectively). In physician-AI test set, the 320 REZ for EfficientNetB0 (AUC: 0.896, P < 0.1) is better than senior physicians. Besides, the 224 REZ for MobileNet (AUC: 0.878, P < 0.1), 448 REZ for Xception (AUC: 0.895, P < 0.1) are better than junior physicians. While the 448 REZ for DenseNet121 (AUC: 0.880, P < 0.05) and ResNet50 (AUC: 0.838, P < 0.05) are only better than entry physicians. ConclusionBased on the gray-scale ultrasound breast images, we obtained the best DL combination which was better than the physicians.

breast cancer deep learning ultrasound resolution artifical intelligence

SCI

英语

第一 ; 通讯

Commission of Science and Technology of Shenzhen[GJHZ20200731095401004]

Oncology

Oncology

WOS:000829754100001

FRONTIERS MEDIA SA

Web of Science

1.Southern Univ Sci & Technol, Jinan Univ, Clin Coll 2, Shenzhen Peoples Hosp,Affiliated Hosp 1,Clin Coll, Shenzhen, Peoples R China
2.Microport Prophecy, Res & Dev Dept, Shanghai, Peoples R China
3.Illuminate LLC Co, Res & Dev Dept, Shenzhen, Peoples R China
4.Chinese Minist Educ, Key Lab Cardiovasc Remodeling & Funct Res, Jinan, Peoples R China
5.Chinese Minist Hlth, Jinan, Peoples R China
6.Shandong Univ, Qilu Hosp, Cheeloo Coll Med, State & Shandong Prov Joint Key Lab Translat Cardi, Jinan, Peoples R China

Wu, Huaiyu,Ye, Xiuqin,Jiang, Yitao,et al. A Comparative Study of Multiple Deep Learning Models Based on Multi-Input Resolution for Breast Ultrasound Images[J]. Frontiers in Oncology,2022,12.

Wu, Huaiyu.,Ye, Xiuqin.,Jiang, Yitao.,Tian, Hongtian.,Yang, Keen.,...&Dong, Fajin.(2022).A Comparative Study of Multiple Deep Learning Models Based on Multi-Input Resolution for Breast Ultrasound Images.Frontiers in Oncology,12.

Wu, Huaiyu,et al."A Comparative Study of Multiple Deep Learning Models Based on Multi-Input Resolution for Breast Ultrasound Images".Frontiers in Oncology 12(2022).