Background
Prostate cancer
Optical microscopy and automated PCa detection
Chemical imaging and automated histologic classification
Overview of this work
Methods
Image Registration
Identification of epithelial cells and their morphologic features
Lumen Detection
Nucleus Detection - single epithelial cells
Feature Extraction
Epithelial cell-related features
Lumen-related features
Global & local tissue features
Feature Selection
Classification
Samples and Data preparation
Results and discussion
The classification system achieves AUC greater than 0.97 on both data sets
Data set | Feature Extraction | AUC | Sensitivity (%) | Specificity (%) | Mf
| ||
---|---|---|---|---|---|---|---|
AVG | STD | AVG | STD | ||||
Data1
| IR & HE | 0.982 | 0.0030 | 90 | 94.76 | 1.64 | 13 |
95 | 90.91 | 1.62 | |||||
99 | 77.80 | 5.52 | |||||
HE only | 0.968 | 0.0052 | 90 | 91.64 | 2.26 | 11 | |
95 | 83.90 | 1.91 | |||||
99 | 53.43 | 13.65 | |||||
Data2
| IR & HE | 0.974 | 0.0145 | 90 | 92.53 | 7.11 | 7 |
95 | 84.19 | 10.84 | |||||
99 | 49.54 | 22.51 | |||||
HE only | 0.880 | 0.0175 | 90 | 61.34 | 10.31 | 8 | |
95 | 22.21 | 10.06 | |||||
99 | 11.21 | 6.01 |
Classification system is robust to staining conditions
Feature Extraction | Data set | AUC | Sensitivity (%) | Specificity (%) | Mf
| ||
---|---|---|---|---|---|---|---|
AVG | STD | AVG | STD | ||||
IR & HE | Train | 0.994 | 0.0006 | 90 | 98.30 | 0.68 | 13 |
95 | 96.58 | 1.10 | |||||
99 | 91.55 | 2.55 | |||||
Test | 0.956 | 0.0089 | 90 | 88.57 | 5.96 | ||
95 | 81.92 | 5.28 | |||||
99 | 26.86 | 15.50 | |||||
HE only | Train | 0.986 | 0.0021 | 90 | 97.77 | 0.97 | 10 |
95 | 91.56 | 2.49 | |||||
99 | 79.29 | 4.47 | |||||
Test | 0.918 | 0.0100 | 90 | 65.51 | 8.37 | ||
95 | 46.14 | 7.53 | |||||
99 | 13.29 | 6.94 |