Train a MULTI-CHANNEL model from scratch on `dcm-zurich-lesions` datasets using nnUNet

[valosekj]

This issue summarizes experiments related to Dataset602_DCMlesionsMultiChannel (multi-channel model trained using T2w_ax and SC seg as input channels to segment lesions).

dataset.json:

    "channel_names": {
        "0": "acq-ax_T2w",
        "1": "SC_seg"
    },
    "labels": {
        "background": 0,
        "lesion": 1
    },

nnUNetPlans.json

```json { "dataset_name": "Dataset602_DCMlesionsMultiChannel", "plans_name": "nnUNetPlans", "original_median_spacing_after_transp": [ 3.5999999046325684, 0.5, 0.5 ], "original_median_shape_after_transp": [ 15, 320, 319 ], "image_reader_writer": "SimpleITKIO", "transpose_forward": [ 0, 1, 2 ], "transpose_backward": [ 0, 1, 2 ], "configurations": { "2d": { "data_identifier": "nnUNetPlans_2d", "preprocessor_name": "DefaultPreprocessor", "batch_size": 31, "patch_size": [ 320, 320 ], "median_image_size_in_voxels": [ 320.0, 319.0 ], "spacing": [ 0.5, 0.5 ], "normalization_schemes": [ "ZScoreNormalization", "ZScoreNormalization" ], "use_mask_for_norm": [ false, false ], "UNet_class_name": "PlainConvUNet", "UNet_base_num_features": 32, "n_conv_per_stage_encoder": [ 2, 2, 2, 2, 2, 2, 2 ], "n_conv_per_stage_decoder": [ 2, 2, 2, 2, 2, 2 ], "num_pool_per_axis": [ 6, 6 ], "pool_op_kernel_sizes": [ [ 1, 1 ], [ 2, 2 ], [ 2, 2 ], [ 2, 2 ], [ 2, 2 ], [ 2, 2 ], [ 2, 2 ] ], "conv_kernel_sizes": [ [ 3, 3 ], [ 3, 3 ], [ 3, 3 ], [ 3, 3 ], [ 3, 3 ], [ 3, 3 ], [ 3, 3 ] ], "unet_max_num_features": 512, "resampling_fn_data": "resample_data_or_seg_to_shape", "resampling_fn_seg": "resample_data_or_seg_to_shape", "resampling_fn_data_kwargs": { "is_seg": false, "order": 3, "order_z": 0, "force_separate_z": null }, "resampling_fn_seg_kwargs": { "is_seg": true, "order": 1, "order_z": 0, "force_separate_z": null }, "resampling_fn_probabilities": "resample_data_or_seg_to_shape", "resampling_fn_probabilities_kwargs": { "is_seg": false, "order": 1, "order_z": 0, "force_separate_z": null }, "batch_dice": true }, "3d_fullres": { "data_identifier": "nnUNetPlans_3d_fullres", "preprocessor_name": "DefaultPreprocessor", "batch_size": 2, "patch_size": [ 16, 320, 320 ], "median_image_size_in_voxels": [ 15.0, 320.0, 319.0 ], "spacing": [ 3.5999979972839355, 0.5, 0.5 ], "normalization_schemes": [ "ZScoreNormalization", "ZScoreNormalization" ], "use_mask_for_norm": [ false, false ], "UNet_class_name": "PlainConvUNet", "UNet_base_num_features": 32, "n_conv_per_stage_encoder": [ 2, 2, 2, 2, 2, 2, 2 ], "n_conv_per_stage_decoder": [ 2, 2, 2, 2, 2, 2 ], "num_pool_per_axis": [ 2, 6, 6 ], "pool_op_kernel_sizes": [ [ 1, 1, 1 ], [ 1, 2, 2 ], [ 1, 2, 2 ], [ 2, 2, 2 ], [ 2, 2, 2 ], [ 1, 2, 2 ], [ 1, 2, 2 ] ], "conv_kernel_sizes": [ [ 1, 3, 3 ], [ 1, 3, 3 ], [ 3, 3, 3 ], [ 3, 3, 3 ], [ 3, 3, 3 ], [ 3, 3, 3 ], [ 3, 3, 3 ] ], "unet_max_num_features": 320, "resampling_fn_data": "resample_data_or_seg_to_shape", "resampling_fn_seg": "resample_data_or_seg_to_shape", "resampling_fn_data_kwargs": { "is_seg": false, "order": 3, "order_z": 0, "force_separate_z": null }, "resampling_fn_seg_kwargs": { "is_seg": true, "order": 1, "order_z": 0, "force_separate_z": null }, "resampling_fn_probabilities": "resample_data_or_seg_to_shape", "resampling_fn_probabilities_kwargs": { "is_seg": false, "order": 1, "order_z": 0, "force_separate_z": null }, "batch_dice": false } }, "experiment_planner_used": "ExperimentPlanner", "label_manager": "LabelManager", "foreground_intensity_properties_per_channel": { "0": { "max": 2102.0, "mean": 575.5780639648438, "median": 516.0, "min": 157.0, "percentile_00_5": 241.0, "percentile_99_5": 1752.0, "std": 235.12808227539062 }, "1": { "max": 1.0, "mean": 1.0, "median": 1.0, "min": 1.0, "percentile_00_5": 1.0, "percentile_99_5": 1.0, "std": 0.0 } } } ```

Dataset602_DCMlesionsMultiChannel will be trained on dcm-zurich-lesions and dcm-zurich-lesions-20231115 datasets using nnUNetv2.

Manual lesion GTs are available for both datasets.

TODO

• [x] convert both datasets from BIDS to nnUNet using dataset_conversion/convert_bids_to_nnUNetv2_multi-channel.py
• [x] run training

[valosekj]

It seems that the model is learning nothing for both fold_0 and fold_1:

fold 0 -- learning nothing -- killed after 500 epochs

fold 1 -- learning nothing -- killed after 500 epochs


[valosekj]

Potential future idea for another MULTI-CHANNEL model:

    "channel_names": {
        "0": "acq-ax_T2w",
        "1": "compression_label"
    },

But we currently do not have the compression labels for dcm-zurich-lesions-20231115.

[valosekj]

Since the multi-channel model with the default nnUNetTrainer trainer was learning nothing (see my comment above), I tried the nnUNetTrainerDiceCELoss_noSmooth trainer (i.e., without the smoothing term of the Dice loss) as done for the region-based model (see here). The training with the nnUNetTrainerDiceCELoss_noSmooth trainer finished successfully 🎉

fold 0, seed42

fold 1, seed42

[valosekj]

Training the multi-channel nnUNetTrainerDiceCELoss_noSmooth model for seed710 now (model name: Dataset603_DCMlesionsMultiChannelSeed710).

[valosekj]

nnUNetTrainerDiceCELoss_noSmooth__nnUNetPlans__3d_fullres/fold_1

`dcm-zurich-lesions`

``` Test Phase Metrics [ANIMA], n=3: Jaccard --> Mean: 0.383, Std: 0.071 Dice --> Mean: 0.549, Std: 0.077 Sensitivity --> Mean: 0.390, Std: 0.079 Specificity --> Mean: 1.000, Std: 0.000 PPV --> Mean: 0.970, Std: 0.043 NPV --> Mean: 1.000, Std: 0.000 RelativeVolumeError --> Mean: -59.436, Std: 9.721 HausdorffDistance --> Mean: 4.537, Std: 2.466 ContourMeanDistance --> Mean: 1.064, Std: 0.241 SurfaceDistance --> Mean: 0.181, Std: 0.193 ```

`dcm-zurich-lesions-20231115`

``` Test Phase Metrics [ANIMA], n=8: Jaccard --> Mean: 0.437, Std: 0.174 Dice --> Mean: 0.587, Std: 0.182 Sensitivity --> Mean: 0.553, Std: 0.293 Specificity --> Mean: 1.000, Std: 0.000 PPV --> Mean: 0.806, Std: 0.128 NPV --> Mean: 1.000, Std: 0.000 RelativeVolumeError --> Mean: -23.972, Std: 53.596 HausdorffDistance --> Mean: 5.372, Std: 5.410 ContourMeanDistance --> Mean: 2.005, Std: 2.353 SurfaceDistance --> Mean: 0.729, Std: 1.423 ```