应用人工智能辅助乳腺癌超声筛查在西藏日喀则地区的初步研究  

Application of artificial intelligence ultrasound in breast cancer screening in Shigatse,Tibet

作　　者：侯永志 田游 白珍 米玛 吴雨林 张迅[2] 常才[2] 周世崇 Hou Yongzhi;Tian You;Bai Zhen;Mi Ma;Wu Yulin;Zhang Xun;Chang Cai;Zhou Shichong(Department of Ultrasound,People′s Hospital of Shigatse City,Shigatse 857000,China;Department of Ultrasound,Fudan University Shanghai Cancer Center&Department of Oncology,Shanghai Medical College,Fudan University,Shanghai 200032,China)

机构地区：[1]日喀则市人民医院超声医学科,日喀则857000 [2]复旦大学附属肿瘤医院超声医学科复旦大学上海医学院肿瘤学系,上海200032

出　　处：《中华超声影像学杂志》2022年第11期927-932,共6页Chinese Journal of Ultrasonography

基　　金：国家自然科学基金(81627804,81830058);上海市科委生物医药科技支撑专项项目(21S31905400);上海市科委国内科技合作项目(21015802300)。

摘　　要：目的:探索人工智能辅助乳腺筛查超声技术在西藏地区乳腺癌筛查中的可行性及临床价值。方法:选择2021年8-9月在日喀则市人民医院参加乳腺癌超声筛查的286例女性入组。研究分为4组:研究组1,高年资乳腺专科超声医师人工超声筛查;研究组2,本地超声医师使用智能辅助超声设备筛查;研究组3,本地超声技师使用智能辅助超声设备筛查;研究组4,本地超声医师人工超声检查。筛查阳性病例的病理结果及阴性病例的6个月超声随访结果设为金标准。结果:21例患者的27个病灶接受活检,病理结果1个浸润性导管癌,1个重度不典型增生,6个纤维腺瘤,5个乳腺病,14个乳腺增生。265例筛查为阴性,6个月超声随访结果依然阴性。研究组2、3、4的准确性、敏感性和特异性分别为0.966、1、0.964,0.935、0.769、0.943,0.860、0.308、0.885。研究组2、3的准确性及曲线下面积与研究组4差异有统计学意义(均P<0.001),与研究组1差异无统计学意义(P=0.063,P=0.055)。结论:人工智能辅助乳腺超声筛查技术可以有效提升非乳腺超声专科医师及技师的筛查效能,非常适合解决西藏地区基层筛查时面临的问题,具有极大的社会意义和临床价值。Objective To explore the feasibility and clinical value of artificial intelligence-assisted breast ultrasound in screening breast cancer in Tibet.Methods Two hundred and eighty-six women who participated in breast cancer screening in Shigatse People′s Hospital from August to September in 2021 were selected.The study included four groups.Group 1,ultrasound screening by senior breast ultrasound doctors from Shanghai;Group 2:local ultrasound doctors used intelligent-assisted ultrasound equipment for screening;Group 3:local ultrasound technicians used intelligent-assisted ultrasound equipment for screening;Group 4:ultrasound screening by local ultrasound doctors.The pathological results of screening positive cases and six-month ultrasound follow-up results of negative cases were set as the gold standard.Results Twenty-seven lesions of 21 persons were screened positive.Pathology showed that 1 case of invasive ductal carcinoma,1 case of severe atypical hyperplasia,6 cases of fibroadenoma,5 cases of breast disease,14 cases of breast hyperplasia.Two hundred and sixty-five persons were screened negative,and the results of the six-month ultrasound follow-up were still negative.The accuracy,sensitivity,and specificity of group 2 were 0.966,1,and 0.964 respectively;The accuracy,sensitivity,and specificity of group 3 were 0.935,0.769,and 0.943 respectively;The accuracy,sensitivity,and specificity of group 4 were 0.860,0.308 and 0.885 respectively.The accuracy and area under the curve of groups 2 and 3 were significantly different from that of group 4(all P<0.001),and there was no significant difference from that of group 1(P=0.063,P=0.055).Conclusions Artificial intelligence-assisted breast ultrasound screening technology can effectively improve the screening efficiency of non-breast ultrasound specialists and technicians.It is very suitable to solve the problems faced by grass-roots screening in Tibet and has great social significance and clinical value.

关 键 词：超声检查 人工智能 乳腺癌筛查 

分 类 号：R445.1[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]