LuGot16 commited on
Commit
0c1f67f
·
verified ·
1 Parent(s): 916957e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -12
README.md CHANGED
@@ -17,16 +17,17 @@ Input 512×512, ImageNet normalization, scale 0.32 µm/px.
17
 
18
  ## 1. Segmentation performance (validation, n = 49)
19
 
20
- | Class | IoU |
21
- |-----------------|-------|
22
- | Background (0) | 0.989 |
23
- | Epithelium (1) | 0.935 |
24
- | Lumen (2) | 0.931 |
25
- | **Mean** | **mIoU 0.952** |
26
 
27
- - Best checkpoint: epoch 47
 
 
28
 
29
- > Metrics reported on the validation set, never on train.
30
 
31
  ![Example segmentation output](https://huggingface.co/LuGot16/Tubules_Segmentation/resolve/main/assets/figure_segmentation_examples_hf.png)
32
 
@@ -106,10 +107,10 @@ Bias = predicted − manual.
106
  and fixation variation.
107
 
108
  **Known biases:**
109
- - Systematic **area** overestimation: ~+12% (tubule) and ~+8% (lumen) vs. manual tracing.
110
- It is systematic (r ≈ 0.99), attributable to the boundary convention (the model mask is
111
- slightly more generous than the manual freehand trace), not noise. It **preserves shape**
112
- (AR/roundness CCC ≈ 0.99). Do not correct with a factor derived from the validation set itself.
113
  - **Lumen perimeter**: good mean agreement (CCC 0.93, bias ≈ 0) but wide LoA (±110 µm):
114
  varies case by case due to tracing convention.
115
 
 
17
 
18
  ## 1. Segmentation performance (validation, n = 49)
19
 
20
+ | Class | IoU | Dice / F1 | clDice |
21
+ |----------------|-------|-----------|--------|
22
+ | Background (0) | 0.990 | 0.995 | — |
23
+ | Epithelium (1) | 0.938 | 0.968 | 0.984 |
24
+ | Lumen (2) | 0.935 | 0.967 | 0.968 |
25
+ | Mean | mIoU 0.954 | 0.977 | — |
26
 
27
+ Per-image (n = 49): mIoU 0.953 ± 0.013 (min 0.900, median 0.955);
28
+ mDice 0.976 ± 0.007 (min 0.946). No image falls below mIoU 0.90.
29
+ Metrics computed on the post-processed predicted masks vs. reference masks; validation set, never train.
30
 
 
31
 
32
  ![Example segmentation output](https://huggingface.co/LuGot16/Tubules_Segmentation/resolve/main/assets/figure_segmentation_examples_hf.png)
33
 
 
107
  and fixation variation.
108
 
109
  **Known biases:**
110
+ - Boundary accuracy vs. reference masks (validation, n = 49):
111
+ - Tubule outer boundary: ASSD 0.59 µm, HD95 1.54 µm, mean signed displacement +0.13 µm (Boundary-F1 @1.6 µm = 0.97).
112
+ - Lumen boundary: ASSD 1.47 µm, HD95 4.99 µm, mean signed displacement −0.05 µm (Boundary-F1 @1.6 µm = 0.75); the larger local variation reflects the convoluted scalloped lumen contour, with no systematic offset.
113
+ The model reproduces the reference masks with sub-pixel mean boundary agreement (signed displacement ≈ 0 for both structures; a +0.13 µm shift accounts for ~0.25% area). The ~+12% / +8% area difference vs. manual ImageJ freehand tracing is therefore a convention difference between the reference masks and the freehand protocol — faithfully inherited by the model — not error introduced by the model. It is systematic (r ≈ 0.99) and preserves shape (AR/roundness CCC ≈ 0.99).
114
  - **Lumen perimeter**: good mean agreement (CCC 0.93, bias ≈ 0) but wide LoA (±110 µm):
115
  varies case by case due to tracing convention.
116