Detection ratios

First, we measure the false positive and missed detection rates on a set of 30 images resulting in a total of more than 5000 fibers, i.e. on half of the images for one fascicle of a nerve. Detected fibers are visually evaluated by super-imposing the edges of the detected fibers on the original image, as illustrated in Figure .

  

Figure: Detected fibers overlaid upon the image of figure

 

Table 4.1: Detection results for the set of 30 images

Image ID	Number of	Detected	Missed	False
	fibers	entities	fibers	positives
1	75	59	16	0
2	143	127	18	2
3	223	181	43	1
4	128	122	9	3
5	171	166	8	3
6	185	173	18	6
7	186	162	26	2
8	194	186	15	7
9	135	126	12	3
10	178	154	30	6
11	230	210	25	5
12	212	181	36	5
13	188	162	28	2
14	189	172	23	6
15	189	172	23	6
16	197	172	31	6
17	195	175	25	5
18	214	196	23	5
19	249	232	26	9
20	247	224	25	2
21	98	95	7	4
22	229	202	36	9
23	233	202	36	9
24	229	206	27	4
25	200	183	21	4
26	204	189	23	8
27	178	165	17	4
28	203	178	29	4
29	190	179	17	6
30	210	187	25	2
Total	5596	5035	691	130

 

For each image, false positives and missed fibers are manually counted. The number of detected entities is determined by the program itself, and the true number of fibers is computed by implementing the correction. The results of this test are shown in Table . The average false positive rate is $2.5 \%$ and the missed detection rate is $11 \%$. This is acceptable for a complete histological study, in which traditionally, information is summarized as histograms of fiber size distribution.