Author Affiliations
1College of Electronic Information Engineering, Hebei University, Baoding, Hebei 0 71002, China2Hebei Software Institute, Baoding, Hebei 0 71030, China3Pathology Department, Baoding Children's Hospital, Baoding, Hebei 0 71051, China4Key Laboratory of Clinical Research on Children's Respiratory Digestive Diseases in Baoding City, Baoding, Hebei 0 71051, Chinashow less
Fig. 1. Pathological section of ANENs
Fig. 2. Overall design flow chart of the system scheme
Fig. 3. Image of ANENs pathological section (left) and color image of YUV (right)
Fig. 4. YUV spatial threshold segmentation images. (a) Threshold segmentation image of positive cell; (b) threshold segmentation image of negative cells
Fig. 5. Images processed by mathematical morphology. (a) Positive cell; (b) negative cells
Fig. 6. One-dimensional minimum highlight process. (a) One-dimensional data; (b) minimum highlight results
Fig. 7. Positive cells processing images obtained by improved watershed algorithm. (a) Image of ANENs pathological section; (b) distance transform image; (c) image by forced minimum technique processing; (d) image processed by improved watershed algorithm; (e) binary segmentation image; (f) color marker segmentation image
Fig. 8. Negative cells processing images obtained by improved watershed algorithm. (a) Image of ANENs pathological section; (b) distance transform images; (c) image by forced minimum technique processing; (d) image processed by improved watershed algorithm; (e) binary segmentation image; (f) color marker segmentation image
Fig. 9. Negative cells segmentation effect diagrams before and after algorithm improvement. (a)--(c) Traditional watershed algorithm; (d)--(f) improved watershed algorithm
Fig. 10. Over-segmentation rate data of negative cells
Fig. 11. Data analysis of Ki-67 index
Fig. 12. Comparison of time consumption on image processing
Cell image No. | Quantity of positive cells | Diagnostic requirement |
---|
Doctor | Traditional algorithm | Improved algorithm |
---|
1 | 10 | 12 | 10 | Yes | 2 | 11 | 13 | 11 | Yes | 3 | 12 | 14 | 12 | Yes | 4 | 9 | 12 | 9 | Yes | 5 | 8 | 11 | 8 | Yes | 6 | 13 | 15 | 12 | Yes | 7 | 11 | 13 | 11 | Yes | 8 | 10 | 12 | 10 | Yes |
|
Table 1. Experimental data of positive cells in ANENs pathological section images obtained by improved watershed algorithm
No. | Negative cell number from doctor | Negative cell number from traditional algorithm | Over-segmentation rate from traditional algorithm /% | Negative cell number from improved algorithm | Over-segmentation rate from improved algorithm /% | Improved algorithm’s accuracy /% |
---|
1 | 423 | 524 | 10.7 | 454 | 3.5 | 92.7 | 2 | 430 | 531 | 10.5 | 465 | 3.9 | 91.9 | 3 | 428 | 516 | 9.3 | 468 | 4.5 | 90.7 | 4 | 370 | 452 | 10.0 | 391 | 2.8 | 94.3 | 5 | 351 | 443 | 11.7 | 368 | 2.4 | 95.2 | 6 | 480 | 576 | 9.1 | 496 | 1.6 | 96.7 | 7 | 421 | 524 | 10.9 | 457 | 4.1 | 91.4 | 8 | 408 | 498 | 9.9 | 430 | 2.6 | 94.6 |
|
Table 2. Experimental data of negative cells in ANENs pathological section images
Experimental group No. | Ki-67 index from doctor /% | Ki-67 index from computer /% | Ki-67 accuracy rate by computer /% | Doctor’s time consumption /s | Computer’s time consumption /s | Error rate from computer /% | Diagnostic value | Diagnostic requirements |
---|
1 | 2.31 | 2.16 | 93.5 | 58 | 29 | 6.5 | <3% | Yes | 2 | 2.50 | 2.31 | 92.4 | 59 | 28 | 7.6 | <3% | Yes | 3 | 2.73 | 2.50 | 91.6 | 60 | 30 | 8.4 | <3% | Yes | 4 | 2.37 | 2.25 | 95.0 | 55 | 32 | 5.0 | <3% | Yes | 5 | 2.23 | 2.12 | 95.0 | 54 | 31 | 5.0 | <3% | Yes | 6 | 2.60 | 2.36 | 90.8 | 58 | 33 | 9.2 | <3% | Yes | 7 | 2.55 | 2.35 | 92.2 | 62 | 27 | 7.8 | <3% | Yes | 8 | 2.39 | 2.27 | 94.9 | 53 | 26 | 5.1 | <3% | Yes |
|
Table 3. Experimental data of diagnostic indicators of ANENs pathological sections