Abstract
Purpose:
Many surgical computer-aided projects ( CAD ) have emerged in recent years, but none interested in cataract surgery . The aim of our study was to develop a method able to recognize in real time the video stream , surgical phases clearly identified and thus to classify them in a database .
Methods:
Over a period of 2 years, we have adopted a first series of 60 videos of cataract surgery , performed by six different surgeons acquired DV format with a resolution of 720x276 pixels. The operating parameters were recorded in postoperative dedicated form. Our first method characterized videos in 9 phases : incision , OVD injection, capsulorhexis , hydrodissection , phacoemulsification , epi - core implementation , OVD aspiration and closure. We developed specifically for this study , a novel characterization method using a granular system at several levels. To recognize in real time during the surgical phase , we used search algorithms for video content. The parameters studied were primarily based on visual content ( movement , shapes, contours , colors, textures ) . Finally , we used the calculation of the area under the ROC curve for the quantification measure of the quality of the results of our method on each of these phases .
Results:
The average video length was 15.25 + / - 7 minutes [ 10.05-28 ] . 8 videos were atypical ( implemented iris hooks , toric IOL) . The mean area under the curve was 0.806 + / - 0.1 [ 0563-0937 ] for the analysis of complete sequences . A gain of 8% was found after improvement of the method using the measurement granularity , which has overcome the overlapping phases and periods of inactivity : 0.887 + / - 0.1 [ 0612-0941 ] .
Conclusions:
We chose to measure granularity in 4 different levels: stage , action, gesture and tool. These recognition systems surgical tasks , whether in level or phases activity level , appear as a significant progress towards the construction of smart systems for surgery. Our method gives good quality results for sequencing our database video cataract surgery. And apart from some interest to have a database by automated grading , we can imagine the development of a device to support the initiation of surgical novice. In this way, real-time applications for intra-operative assistance may be developped , for example by allowing real time to know which information needs to be showned to the surgeon for the task performed ..
Keywords: 549 image processing •
445 cataract •
713 shape, form, contour, object perception