实验室参加ImageCLEF2016 Medical Task评测取得第一名
新闻来源:IR实验室       发布时间:2016/8/23 17:59:01

近日,实验室博士生于玉海和赵哲焕参加了2016年的ImageCLEF Medical Task评测(简称ImageCLEF2016med),并取得优异成绩。本评测由国际组织The CLEF Initiative主办,吸引了来自全球学术界和工业界的广泛关注和参与,本次评测任务旨在识别、标注和分割生物医学文献中的复合图像,为后续子图模式分类和子图理解奠定基础,并最终对医学教育、研究和临床诊断提供服务。

ImageCLEFmed 2016吸引了来自工业界和学术界共72个参赛队伍,其中我们实验室的DUTIR、美国特拉华大学的CIS UDEL和希腊亚里士多德大学的MKLD三个参赛队伍,参加本次评测的子任务:复合图像探测(Compound Figure Detection)。在所有提交的15个提交的结果中,我们DUTIR参赛队获得第名的成绩,比第二名参赛队结果高出接近2个百分点(详见http://www.imageclef.org/2016/medical)。

1.png

                                             

根据评测结果,我们进一步改进模型并撰写论文《Assembling Deep Neural Networks for Medical Compound Figure Detection》,论文摘要如下:

 

Abstract Compound figure detection on figures and associated captions is the first step to make medical figures from biomedical literature available for further analysis. The performance of traditional hand-engineering methods is limited to the choice of features and the prior domain knowledge. We train multiple Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRU) networks on top of pre-trained word vectors to learn textual features from captions and employ deep Convolutional Neural Networks (CNN) to extract visual features from figures. We then identify compound figures by combining textual and visual prediction. Our proposed architecture obtains remarkable performance in three run types of textual, visual and mixed and achieves new state-of-the-art accuracies of 88.07% in ImageCLEF2015 and 96.24% in ImageCLEF2016.

Keywords: compound figure detection; convolutional neural network; recurrent neural network; word vectors;