您好

[111678900]

您好，请问计算分割准确性，包括灵敏度，特异性，Hausdorff距离95%，平均表面距离和残余均方距离。代码该怎么写？

[TaWald]

Can you please provide a translation for your question in English so it is more accesible for other users.?

[111678900]

Hello, I apologize for the inconvenience. Do you have a function developed for calculating segmentation accuracy, including sensitivity, specificity, Hausdorff distance 95%, average surface distance, and residual mean square distance?

[111678900]

I have completed the training process. I'm wondering if you have developed functionality for calculating segmentation accuracy, including sensitivity, specificity, Hausdorff distance 95%, average surface distance, and residual mean square distance. Thank you very much.

[e-FunJs]

找到.../nnunetv2/evaluation/evaluate_predictions.py这个文件，在里面添加函数。nnUNetv1有这些指标，但是没法仿照着写。建议自己写吧？ 这是我的，不一定对： ...

        results['metrics'][r]['Sensitivity'] = tp / (tp + fn) if (tp + fn) > 0 else np.nan
        # Sensitivity = TP/TP+FN (Recall)
        results['metrics'][r]['Specificity'] = tn / (tn + fp) if (tn + fp) > 0 else np.nan
        # Specificity = TN/TN+FP
        results['metrics'][r]['Jaccard'] = tp / (tp + fp + fn) if (tp + fp + fn) != 0 else np.nan
        # Jaccard = TP/TP+FP+FN
        #Hausdorff distance
        results['metrics'][r]['hausdorff distance'] = hd
        results['metrics'][r]['hausdorff distance 95'] = hd95

...

def compute_hausdorff(test, reference):
    # test & reference: shape(1, 155, 240, 240)
    test_slices = test[0]  # (155, 240, 240)
    reference_slices = reference[0] # (155, 240, 240)

    all_hd = []

    for i in range(test_slices.shape[0]):  
        slice_test = test_slices[i]
        slice_reference = reference_slices[i]
        if slice_test.any() and slice_reference.any():  
            hd = max(directed_hausdorff(slice_test, slice_reference)[0],
                     directed_hausdorff(slice_reference, slice_test)[0])
            all_hd.append(hd)
    if all_hd:
        hd95 = np.percentile(all_hd, 95)
        return max(all_hd), hd95
    else:
        return np.nan, np.nan  

[TaWald]

I assume this issue is concluded with @e-FunJs answer