Skip to main content

Table 2 Performance of the proposed de-arraying method on three datasets under four setting options: (1) both the segmentation and the non-linear estimation (for the deformation) modules are deactivated, (2) the segmentation is activated but the non-linear estimation is deactivated, (3) the segmentation is deactivated and the non-linear estimation is activated, and (4) both of them are activated

From: ATMAD: robust image analysis for Automatic Tissue MicroArray De-arraying

  Localization   Estimation of array coordinates  
  A (a) P (b) R (c) F (d) G (e) JSC (f) A (a) P (b) R (c) F (d) G (e) JSC (f)
SIMULATED IMAGES  
Average on 31 images  
Wang et al. [22] 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
ATMAD Opt. #1 1 1 1 1 1 1 0.93 1 0.93 0.96 0.96 0.93
Opt. #2 0.95 1 0.95 0.98 0.98 0.95 0.85 1 0.84 0.91 0.91 0.84
Opt. #3 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #4 0.95 1 0.95 0.98 0.98 0.95 0.95 1 0.95 0.98 0.98 0.95
Fig. 11a  
ATMAD Opt. #1 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #2 0.97 1 0.96 0.98 0.98 0.96 0.97 1 0.96 0.98 0.98 0.96
Opt. #3 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #4 0.97 1 0.96 0.98 0.98 0.96 0.97 1 0.96 0.98 0.98 0.96
Fig. 11b  
ATMAD Opt. #1 1 1 1 1 1 1 0.93 1 0.92 0.96 0.96 0.92
Opt. #2 0.93 1 0.92 0.96 0.96 0.92 0.90 1 0.89 0.94 0.95 0.89
Opt. #3 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #4 0.93 1 0.92 0.96 0.96 0.92 0.93 1 0.92 0.96 0.96 0.92
BRIGHTFIELD IMAGES  
Average on 8 images  
ATMAD Opt. #1 0.94 1 0.93 0.96 0.96 0.93 0.91 1 0.88 0.94 0.94 0.88
Opt. #2 0.87 1 0.83 0.91 0.91 0.83 0.84 1 0.79 0.88 0.89 0.79
Opt. #3 0.94 1 0.93 0.96 0.96 0.93 0.94 1 0.93 0.96 0.96 0.93
Opt. #4 0.87 1 0.83 0.91 0.91 0.83 0.93 1 0.91 0.95 0.95 0.91
Fig. 12  
ATMAD Opt. #1 0.96 1 0.94 0.97 0.97 0.94 0.91 1 0.89 0.94 0.94 0.89
Opt. #2 0.87 1 0.83 0.91 0.91 0.83 0.84 1 0.80 0.89 0.89 0.80
Opt. #3 0.96 1 0.94 0.97 0.97 0.94 0.96 1 0.94 0.97 0.97 0.94
Opt. #4 0.87 1 0.83 0.91 0.91 0.83 0.93 1 0.91 0.95 0.95 0.91
FLUORESCENCE IMAGES  
Average on 8 DNA microarray images  
ATMAD Opt. #1 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #2 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #3 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #4 1 1 1 1 1 1 1 1 1 1 1 1
Fig. 13  
ATMAD Opt. #1 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #2 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #3 1 1 1 1 1 1 1 1 1 1 1 1
Opt. #4 1 1 1 1 1 1 1 1 1 1 1 1
Average on 4 TMA images  
ATMAD Opt. #1 0.79 0.77 1 0.87 0.88 0.77 0.92 0.98 0.93 0.95 0.95 0.91
Opt. #2 0.91 0.99 0.90 0.94 0.95 0.89 0.73 1 0.69 0.82 0.83 0.69
Opt. #3 0.79 0.77 1 0.87 0.88 0.77 - - - - - -
Opt. #4 0.91 0.99 0.90 0.94 0.95 0.89 0.92 0.98 0.93 0.96 0.96 0.91
Fig. 14  
ATMAD Opt. #1 0.88 0.86 1 0.93 0.93 0.86 0.94 1 0.94 0.97 0.97 0.94
Opt. #2 0.86 1 0.84 0.91 0.92 0.84 0.80 1 0.77 0.87 0.88 0.77
Opt. #3 0.88 0.86 1 0.93 0.93 0.86 0.97 1 0.97 0.98 0.98 0.97
Opt. #4 0.86 1 0.84 0.91 0.92 0.84 1 1 1 1 1 1
  1. All considered performance scores range from 0 (worst) to 1 (best) and measure the similarity between the de-arraying results and their corresponding ground-truth or manual annotation. Notations: (a) accuracy (A), (b) precision (P), (c) recall (R), (d) F-score (F), (e) G-score (G) and (f) Jaccard coefficient (JSC)