英文是纯手打的!论文原文的summarizing and paraphrasing。可能会出现难以避免的拼写错误和语法错误,若有发现欢迎评论指正!文章偏向于笔记,谨慎食用
目录
2.3.2. Real-Time Pupil Recognition and Phase Classification
2.3.5. Platform Performance Analysis
2.3.6. Surgical Usability Evaluation
2.4.3. Final Integrated Surgical Guidance Solution
1. 心得
(1)好神奇的期刊名字。怎么感觉看doi还没有简称的样子,我应该咋写标题?
(2)还有视频!?点开之后眼睛已经开始隐隐作痛了!!开始觉得医生有点太厉害了(不会医生也觉得我们做AI的很不可思议吧)
(3)每次到了新交叉领域要猛猛学英语!!!我的舌头要打架了
(4)好像还挺有用的
2. 论文逐段精读
2.1. Abstract
①Aiming: pupil tracking, surgical phase identifying and real-time visual feedback in phacoemulsification
intraoperative adj. 手术中的 cataract n.白内障;大瀑布
phacoemulsification n. (白内障的)超声乳化术
pupil n. 学生;瞳孔;(尤指)小学生;弟子;眸子;门生;瞳人(天哪高中都当学生背的,今天第一次看到用作瞳孔的(虽然知道有这个叫法,但以为会很偏门那种))
ophthalmology n.眼科学 capsulorhexis n.撕囊;晶状体撕囊术;撕囊术
2.2. Introduction
①Variables that influence the result of opeartion: rapid changes in intracameral flow and associated fluidic turbulence, instrument positioning and proximity to tissues such as the posterior lens capsule, and visualization of intraocular tissues
②Tasks of DNN: capsulorhexis guiding, feedback on decentration of the eye, erratic tool movement, and turbulent flow conditions, anatomical visualization enhancement
intracameral adj. 眼内镜的 erratic adj.不稳定的;不规则的;不确定的;不可靠的 n. 漂砾;漂石
rhexis n.破裂
2.3. Methods
2.3.1. Data Set
①Equipment: NGENUITY 3D Visualization System; Alcon
②Surgical procedures: 6
③Frame selection: 600
④600 frames from 6 eyes are training set, 200 for 2 eyes are val and left 2 for test
⑤Evaluation dataset: publicly available Cataract-101 data set, with 101 phacoemulsification cataract procedures and 10 100 frames
stereoscopic adj.有立体视觉的;有立体效果的
ocular adj. 眼的;眼睛的;看得到的;可以看见的 n.〈谑〉眼睛;目镜
2.3.2. Real-Time Pupil Recognition and Phase Classification
①Model: Faster R-CNN based ResNet-50
②Tasks: stage classification and pupil detection
2.3.3. Data Pipeline
①Each 5 frames are send to Faster R-CNN?
②Pipeline of model:
2.3.4. Computer Vision Tools
①Tasks: exposed capsular bag classification, optical flow tracking, pupil decentered detection, rhexis path prediction(怎么我读完这段话感觉Faster R-CNN做了好多件事情?功能这么齐全吗?翻了补充文件好像就看到前两个说是调库)
limbus n.边缘;异色边缘,异组织边缘
2.3.5. Platform Performance Analysis
①For phase classification: AUROC and AUPR
②For pupil segmentation: intersection over union, precision, recall, Dice score of the segmented area vs the annotator’s area of the pupil, and the percentage of capsular bag exposed compared with the expert’s opinion(还有人工的!!工作量挺大啊)
③Pupil area annotation:
where the red box is predicted by model and the blue one is annotated by experts
2.3.6. Surgical Usability Evaluation
~
2.4. Results
2.4.1. DNN Predictions
①Performance of phase classification:
(精度好高啊,四分类?)
②Other performance:
③Overall agreement among annotators:
④Classification pipeline:
2.4.2. Computer Vision Tools
①怎么又来这个标题的章节??
②Discrepancy of capsular bag:
③Agreement consistency between annotators only:
④为什么K means组群=2但是分出来了三个东西?边框为什么没有被分进去?
2.4.3. Final Integrated Surgical Guidance Solution
①“手术引导平台在整个囊切术的整个执行过程中显示了一个具有可定制直径的圆形 rhexis 模板(A)。光流功能计算器械的加速度和晶状体碎片的移动,当加速度值达到外科医生设定的阈值时,增加外圆厚度(B)。当DNN模型识别皮层纤维的去除时,该平台应用了相同的光流跟踪工具(C)。当瞳孔中未检测到任何器械时,空闲阶段的分配会停用所有手术引导工具(D)”
②FPS: 137/80/92 for 3 phases
③Post hoc evaluation:
2.5. Discussion
~
2.5.1. Limitations
①Generalization
2.6. Conclusions
~