Video Retrieval System for Meniscal Surgery to Improve Health Care Services

Table 2

Comparative analysis of proposed system with already existing systems.

	Dataset	Clip size	Average retrieval time	Average number of match frames/videos	Recall	Average precision	ROC	Description
1	Medical videos & classified human action	Avg 20 sec	7 min 3 sec	0	0%	0.43	0	Time required to compute feature vector for 9 videos; 0.43 in case of ERM and CD dataset
2	Medical videos	1.97 min and above	16.4 (retrieval time for all attributes)	111	0%	0.983	0	Precision is based on 8 videos using the neural network in terms of relevant key frame retrieval and a total number of key frames for that video
3	Laparoscopic videos		0	0	80%	0	0	N/A
4	Endoscopic videos	60 sec to 250 sec	0	0	88%	0	0	N/A
5	Endoscopy	0	0	0	0%	0.78	0	N/A
6	Cataract	0	0	0	0%		0.7	With motion analysis and 0.9 with the presence of surgical process
7	Hysteroscopy	2 to 3 min	0	0	0%	0.92	0
8	Eye surgery	0	0	0	0%	0.72	0	In the case of CSD dataset
9	Epiretinal surgery	0	0	0	0%	0.62	0	0.69 in case of motion histogram
Proposed	Meniscal surgery	60 sec	60 sec	10	78%	0.57; 0.68; 0.64; 0.78	0	0.78 for PCA-SIFT