Table 8 Performance comparison of TR-MAMIL using different feature extractor methods on the testing sets in (a) classification of aggressive and non-aggressive EC, (b) TMB prediction for the aggressive EC, and (c) TMB prediction for the non-aggressive EC

(a) Classification of aggressive and non-aggressive EC

Truncated ResNet50

F1-score

0.89

0.93

0.84

0.89

0.97

Truncated ResNet101

0.84

0.80

0.89

0.84

0.86

Truncated ResNet152

0.86

0.83

0.90

0.87

0.93

CCL-ResNet50⁶³

0.58

1.00

0.00

0.50

0.53

CCL-ResNet152

0.58

1.00

0.00

0.50

0.52

DINO-ViT-S/16⁶⁴

0.84

0.90

0.77

0.83

0.90

Lunit’s DINO-ViT-S/16⁵⁸

0.86

0.82

0.90

0.86

0.91

Lunit’s DINO-ViT-S/16 with data normalization⁵⁸

0.85

0.83

0.87

0.85

0.91

(b) TMB prediction for aggressive ECs

Truncated ResNet50

Cross-Entropy

0.67

0.78

0.61

0.69

0.69

Truncated ResNet101

0.71

0.75

0.68

0.71

0.78

Truncated ResNet152

0.73

0.63

0.73

0.73

0.78

CCL-ResNet50⁶³

0.58

1.00

0.00

0.50

0.50

CCL-ResNet152

0.58

1.00

0.00

0.50

0.48

DINO-ViT-S/16⁶⁴

0.65

0.83

0.54

0.69

0.70

Lunit’s DINO-ViT-S/16⁵⁸

0.70

0.75

0.66

0.71

0.76

Lunit’s DINO-ViT-S/16 with data normalization⁵⁸

0.72

0.69

0.73

0.71

0.79

(c) TMB prediction for non-aggressive ECs

Truncated ResNet50

F1-score

0.64

0.57

0.67

0.62

0.69

Truncated ResNet101

0.69

0.53

0.84

0.61

0.65

Truncated ResNet152

0.66

0.76

0.61

0.68

0.70

CCL-ResNet50⁶³

0.58

1.00

0.00

0.50

0.67

CCL-ResNet152

0.58

1.00

0.00

0.50

0.43

DINO-ViT-S/16⁶⁴

0.59

0.43

0.67

0.55

0.51

Lunit’s DINO-ViT-S/16⁵⁸

0.48

0.57

0.44

0.51

0.51

Lunit’s DINO-ViT-S/16 with data normalization⁵⁸

0.55

0.38

0.63

0.50

0.58