Ann Transl Med:人工智能检测白癜风病变能代替医生的观察吗?

2022-07-27 医路坦克 MedSci原创

该研究是为了建立和验证一个基于深度学习的混合人工智能(AI)模型,以客观地评估白癜风病变的形态和比色。并评估AI模型与皮肤科医生评估者之间的一致性。

该研究为了建立和验证一个基于深度学习的混合人工智能(AI)模型,以客观地评估白癜风病变的形态和比色。评估AI模型与皮肤科医生评估者之间的一致性。

 建立了中国白癜风患者(Fitzpatrick皮肤类型III或IV)白癜风病变图像的两个主要数据集,一个有2720张图像用于病灶定位研究,一个有1262张图像用于病灶分割研究。此外,还生成了一个包含145张来自其他Fitzpatrick皮肤类型(I、II或V)的白癜风病变图像的额外测试集。构建了三级混合模型。训练并验证了YOLO v3 (Y ou Only Look Once, v3)架构,以敏感性和错误率作为主要性能指标,对白癜风病灶进行分类和定位。然后基于Jaccard指数(JI)对3种深度卷积神经网络(DCNNs)、金字塔场景解析网络(PSPNet)、UNet和UNet++进行了分割研究。将性能最好的体系结构集成到模型中。最后开发了三个附加指标,即VAreaA、VAreaR和VColor,分别用于测量绝对大小变化、相对大小变化和色素沉着。

图1 我们基于dcnn的混合AI模型的开发和验证概述。该图像的发布得到了患者或患者法定监护人的同意。DCNN,深度卷积神经网络;人工智能,人工智能;VAreaA,一个评估绝对病变大小的指标;VAreaR,评估相对尺寸变化的指标;VColor,评价比色(色素沉着)变化的指标;PS, Photoshop。

表1    比较模型的相对尺寸变化测量,Photoshop分析,和皮肤科医生

表2    对模型和皮肤科医生的比色评估结果进行T检验

YOLO v3架构检测白癜风病灶的灵敏度为92.91%,错误率为14.98%。UNet++体系结构在分割研究中优于其他体系结构(JI, 0.79),并被集成到模型中。而在附加测试集上,模型的检测灵敏度较低(72.41%),分割得分较低(JI, 0.69)。在尺寸变化方面,AI模型与训练过的皮肤科医生(W=0.812, P<0.05)、Photoshop分析(P=0.075, P=0.212)无差异,均表现出良好的一致性。

我们开发了一种新颖、方便、客观、定量的基于深度学习的混合模型,该模型可以同时评估III或IV型Fitzpatrick皮肤患者的白癜风病变形态和比色,因此该模型适用于临床和研究场景中评估亚洲人白癜风病变的严重程度。在其他种族的皮肤群体中使用它也需要更多的工作。

文献来源:Guo L,  Yang Y,  Ding H,A deep learning-based hybrid artificial intelligence model for the detection and severity assessment of vitiligo lesions.Ann Transl Med 2022 May;10(10)

 

版权声明:
本网站所有内容来源注明为“梅斯医学”或“MedSci原创”的文字、图片和音视频资料,版权均属于梅斯医学所有。非经授权,任何媒体、网站或个人不得转载,授权转载时须注明来源为“梅斯医学”。其它来源的文章系转载文章,或“梅斯号”自媒体发布的文章,仅系出于传递更多信息之目的,本站仅负责审核内容合规,其内容不代表本站立场,本站不负责内容的准确性和版权。如果存在侵权、或不希望被转载的媒体或个人可与我们联系,我们将立即进行删除处理。
在此留言
评论区 (4)
#插入话题
  1. [GetPortalCommentsPageByObjectIdResponse(id=2067635, encodeId=a38d206e6352e, content=<a href='/topic/show?id=d0841e75535' target=_blank style='color:#2F92EE;'>#Transl#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=61, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17755, encryptionId=d0841e75535, topicName=Transl)], attachment=null, authenticateStatus=null, createdAvatar=https://thirdqq.qlogo.cn/qqapp/101173734/363C99735735B7F4CB11D5F78F580525/100, createdBy=aab22500211, createdName=ms3025846204744509, createdTime=Wed May 17 12:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1657661, encodeId=4b67165e6611f, content=<a href='/topic/show?id=3ca31e71903' target=_blank style='color:#2F92EE;'>#TRA#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=75, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17719, encryptionId=3ca31e71903, topicName=TRA)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=757b24868428, createdName=bsmagic9140, createdTime=Wed May 17 02:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1234992, encodeId=07b9123499289, content=学习了, beContent=null, objectType=article, channel=null, level=null, likeNumber=74, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=70778248916, createdName=张得帅, createdTime=Wed Jul 27 08:08:18 CST 2022, time=2022-07-27, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1611673, encodeId=b1e216116e318, content=<a href='/topic/show?id=eb6b1145939' target=_blank style='color:#2F92EE;'>#Med#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=59, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=11459, encryptionId=eb6b1145939, topicName=Med)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=552a19396740, createdName=ms3994565386320060, createdTime=Thu Jul 21 13:36:28 CST 2022, time=2022-07-21, status=1, ipAttribution=)]
  2. [GetPortalCommentsPageByObjectIdResponse(id=2067635, encodeId=a38d206e6352e, content=<a href='/topic/show?id=d0841e75535' target=_blank style='color:#2F92EE;'>#Transl#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=61, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17755, encryptionId=d0841e75535, topicName=Transl)], attachment=null, authenticateStatus=null, createdAvatar=https://thirdqq.qlogo.cn/qqapp/101173734/363C99735735B7F4CB11D5F78F580525/100, createdBy=aab22500211, createdName=ms3025846204744509, createdTime=Wed May 17 12:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1657661, encodeId=4b67165e6611f, content=<a href='/topic/show?id=3ca31e71903' target=_blank style='color:#2F92EE;'>#TRA#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=75, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17719, encryptionId=3ca31e71903, topicName=TRA)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=757b24868428, createdName=bsmagic9140, createdTime=Wed May 17 02:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1234992, encodeId=07b9123499289, content=学习了, beContent=null, objectType=article, channel=null, level=null, likeNumber=74, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=70778248916, createdName=张得帅, createdTime=Wed Jul 27 08:08:18 CST 2022, time=2022-07-27, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1611673, encodeId=b1e216116e318, content=<a href='/topic/show?id=eb6b1145939' target=_blank style='color:#2F92EE;'>#Med#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=59, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=11459, encryptionId=eb6b1145939, topicName=Med)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=552a19396740, createdName=ms3994565386320060, createdTime=Thu Jul 21 13:36:28 CST 2022, time=2022-07-21, status=1, ipAttribution=)]
    2023-05-17 bsmagic9140
  3. [GetPortalCommentsPageByObjectIdResponse(id=2067635, encodeId=a38d206e6352e, content=<a href='/topic/show?id=d0841e75535' target=_blank style='color:#2F92EE;'>#Transl#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=61, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17755, encryptionId=d0841e75535, topicName=Transl)], attachment=null, authenticateStatus=null, createdAvatar=https://thirdqq.qlogo.cn/qqapp/101173734/363C99735735B7F4CB11D5F78F580525/100, createdBy=aab22500211, createdName=ms3025846204744509, createdTime=Wed May 17 12:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1657661, encodeId=4b67165e6611f, content=<a href='/topic/show?id=3ca31e71903' target=_blank style='color:#2F92EE;'>#TRA#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=75, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17719, encryptionId=3ca31e71903, topicName=TRA)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=757b24868428, createdName=bsmagic9140, createdTime=Wed May 17 02:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1234992, encodeId=07b9123499289, content=学习了, beContent=null, objectType=article, channel=null, level=null, likeNumber=74, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=70778248916, createdName=张得帅, createdTime=Wed Jul 27 08:08:18 CST 2022, time=2022-07-27, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1611673, encodeId=b1e216116e318, content=<a href='/topic/show?id=eb6b1145939' target=_blank style='color:#2F92EE;'>#Med#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=59, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=11459, encryptionId=eb6b1145939, topicName=Med)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=552a19396740, createdName=ms3994565386320060, createdTime=Thu Jul 21 13:36:28 CST 2022, time=2022-07-21, status=1, ipAttribution=)]
    2022-07-27 张得帅

    学习了

    0

  4. [GetPortalCommentsPageByObjectIdResponse(id=2067635, encodeId=a38d206e6352e, content=<a href='/topic/show?id=d0841e75535' target=_blank style='color:#2F92EE;'>#Transl#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=61, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17755, encryptionId=d0841e75535, topicName=Transl)], attachment=null, authenticateStatus=null, createdAvatar=https://thirdqq.qlogo.cn/qqapp/101173734/363C99735735B7F4CB11D5F78F580525/100, createdBy=aab22500211, createdName=ms3025846204744509, createdTime=Wed May 17 12:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1657661, encodeId=4b67165e6611f, content=<a href='/topic/show?id=3ca31e71903' target=_blank style='color:#2F92EE;'>#TRA#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=75, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=17719, encryptionId=3ca31e71903, topicName=TRA)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=757b24868428, createdName=bsmagic9140, createdTime=Wed May 17 02:36:28 CST 2023, time=2023-05-17, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1234992, encodeId=07b9123499289, content=学习了, beContent=null, objectType=article, channel=null, level=null, likeNumber=74, replyNumber=0, topicName=null, topicId=null, topicList=[], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=70778248916, createdName=张得帅, createdTime=Wed Jul 27 08:08:18 CST 2022, time=2022-07-27, status=1, ipAttribution=), GetPortalCommentsPageByObjectIdResponse(id=1611673, encodeId=b1e216116e318, content=<a href='/topic/show?id=eb6b1145939' target=_blank style='color:#2F92EE;'>#Med#</a>, beContent=null, objectType=article, channel=null, level=null, likeNumber=59, replyNumber=0, topicName=null, topicId=null, topicList=[TopicDto(id=11459, encryptionId=eb6b1145939, topicName=Med)], attachment=null, authenticateStatus=null, createdAvatar=null, createdBy=552a19396740, createdName=ms3994565386320060, createdTime=Thu Jul 21 13:36:28 CST 2022, time=2022-07-21, status=1, ipAttribution=)]

相关资讯

European Radiology:人工智能如何改善影像科医生的肺栓塞诊断?

在2019年冠状病毒病(COVID-19)大流行期间,全球PE发病率增加了2倍多。

European Radiology:人工智能,让乳腺MR非肿块样增强病变的诊断更有底气!

人工智能(AI)是一项可以从医学图像中提取比放射科医生进行的视觉分析多得多的特征来建立计算机辅助诊断系统。

Med Image Anal:国际领先!中山六院实现人工智能结肠癌病灶影像精准分割

当前,能否实现全自动的、精准的病灶分割是决定医学影像在临床使用效果的关键。

Nature子刊:基于AI的预测模型,可快速发现基因组中的致癌突变

这些发现表明,常见驱动突变与罕见驱动突变之间实际上有很多重叠之处。

Chin Med:开发中西医结合中肝毒性化合物的人工智能筛选方法

中西医结合(TCM-WMC)增加了摄入的化合物的复杂性。利用人工智能(AI)方法开发一种基于化学结构的中药白酒中肝毒性化合物的筛选方法。

DeepMind再发重磅论文:让计算机像婴儿一样思考

年幼的婴儿也懂得“直观物理”——即世界运作的常识性规则。只有5个月大的婴儿在看到了不符合物理规律的场景也会感到惊讶,比如玩具突然不见了。