Merge pull request #19 from ivadomed/um/ivadomed_scseg_training

`ivadomed` configs for training spinal cord segmentation

config/seg_sc.json

@@ -0,0 +1,148 @@
+{
+    "command": "train",
+    "gpu_ids": [0],
+    "path_output": "seg_sc_output",
+    "model_name": "seg_sc_model",
+    "debugging": true,
+    "object_detection_params": {
+        "object_detection_path": null,
+        "safety_factor": [1.0, 1.0, 1.0]
+    },
+    "loader_parameters": {
+        "path_data": ["basel-mp2rage-preprocessed/data_processed_scseg"],
+        "subject_selection": {
+            "n": [],
+            "metadata": [],
+            "value": []
+        },
+        "target_suffix": ["_seg-manual"],
+        "extensions": [],
+        "roi_params": {
+            "suffix": null,
+            "slice_filter_roi": null
+        },
+        "contrast_params": {
+            "training_validation": ["UNIT1"],
+            "testing": ["UNIT1"],
+            "balance": {}
+        },
+        "slice_filter_params": {
+            "filter_empty_mask": false,
+            "filter_empty_input": false
+        },
+        "slice_axis": "sagittal",
+        "multichannel": false,
+        "soft_gt": false
+    },
+    "split_dataset": {
+        "fname_split": null,
+        "random_seed": 42,
+        "center_test": [],
+        "method": "per_patient",
+        "balance": null,
+        "train_fraction": 0.75,
+        "test_fraction": 0.2
+    },
+    "training_parameters": {
+        "batch_size": 2,
+        "loss": {
+            "name": "DiceLoss"
+        },
+        "training_time": {
+            "num_epochs": 200,
+            "early_stopping_patience": 50,
+            "early_stopping_epsilon": 0.001
+        },
+        "scheduler": {
+            "initial_lr": 1e-3,
+            "lr_scheduler": {
+                "name": "CosineAnnealingLR",
+                "base_lr": 1e-5,
+                "max_lr": 1e-3
+            }
+        },
+        "balance_samples": {
+            "applied": false,
+            "type": "gt"
+        },
+        "mixup_alpha": null,
+        "transfer_learning": {
+            "retrain_model": null,
+            "retrain_fraction": 1.0,
+            "reset": true
+        }
+    },
+    "default_model": {
+        "name": "Unet",
+        "dropout_rate": 0.3,
+        "bn_momentum": 0.1,
+        "depth": 4,
+        "is_2d": false,
+        "final_activation": "sigmoid"
+    },
+    "uncertainty": {
+        "epistemic": false,
+        "aleatoric": false,
+        "n_it": 0
+    },
+    "postprocessing": {"binarize_prediction": {"thr": 0.5}},
+    "evaluation_parameters": {},
+    "transformation": {
+        "Resample": {
+            "wspace": 1.0,
+            "hspace": 1.0,
+            "dspace": 1.0
+        },
+        "CenterCrop": {
+            "size": [256, 256, 256]
+        },
+        "RandomReverse": {
+            "applied_to": ["im", "gt"],
+            "dataset_type": ["training"]
+        },
+        "RandomAffine": {
+            "degrees": 10,
+            "scale": [0.2, 0.2, 0.2],
+            "translate": [0.0, 0.0, 0.0],
+            "applied_to": ["im", "gt"],
+            "dataset_type": ["training"]
+        },
+        "ElasticTransform": {
+            "alpha_range": [25.0, 35.0],
+            "sigma_range": [3.5, 5.5],
+            "p": 0.5,
+            "applied_to": ["im", "gt"],
+            "dataset_type": ["training"]
+        },
+        "RandomGamma": {
+            "log_gamma_range": [-1.0, 1.0],
+            "p": 0.5,
+            "applied_to": ["im"],
+            "dataset_type": ["training"]
+        },
+        "RandomBiasField": {
+            "coefficients": 0.5,
+            "order": 3,
+            "p": 0.3,
+            "applied_to": ["im"],
+            "dataset_type": ["training"]
+        },
+        "RandomBlur": {
+            "sigma_range": [0.0, 2.0],
+            "p": 0.3,
+            "applied_to": ["im"],
+            "dataset_type": ["training"]
+        },
+        "NumpyToTensor": {},
+        "NormalizeInstance": {
+            "applied_to": ["im"]
+        }
+    },
+    "Modified3DUNet": {
+        "applied": true,
+        "length_3D": [256, 256, 256],
+        "stride_3D": [256, 256, 256],
+        "attention": false,
+        "n_filters": 8
+    }
+}

preprocessing/qc_preprocess.py

@@ -30,7 +30,7 @@
 subjects = subjects_df['participant_id'].values.tolist()
 
 # Log resolutions and sizes for data exploration
-resolutions, sizes = [], []
+resolutions, sizes, crop_sizes = [], [], []
 
 # Log problematic subjects for QC
 failed_crop_subjects, shape_mismatch_subjects, left_out_lesion_subjects = [], [], []
@@ -41,18 +41,22 @@
     subject_images_path = os.path.join(args.sct_output_path, 'data_processed', subject, 'anat')
     subject_labels_path = os.path.join(args.sct_output_path, 'data_processed', 'derivatives', 'labels', subject, 'anat')
 
-    # Read cropped subject image (i.e. 3D volume) to be used for training
+    # Read original and cropped subject image (i.e. 3D volume) to be used for training
+    img_path = os.path.join(subject_images_path, '%s_UNIT1.nii.gz' % subject)
     img_crop_fpath = os.path.join(subject_images_path, '%s_UNIT1_crop.nii.gz' % subject)
     if not os.path.exists(img_crop_fpath):
         failed_crop_subjects.append(subject)
         continue
+    img = nib.load(img_path)
     img_crop = nib.load(img_crop_fpath)
 
     # Get and log size and resolution for each subject image
-    size = img_crop.get_fdata().shape
+    size = img.get_fdata().shape
+    crop_size = img_crop.get_fdata().shape
     resolution = tuple(img_crop.header['pixdim'].tolist()[1:4])
     resolution = tuple([np.round(r, 1) for r in list(resolution)])
     sizes.append(size)
+    crop_sizes.append(crop_size)
     resolutions.append(resolution)
 
     # Read original and cropped subject ground-truths (GT)
@@ -78,6 +82,7 @@
 
 print('RESOLUTIONS: ', Counter(resolutions))
 print('SIZES: ', Counter(sizes))
+print('CROP SIZES: ', Counter(crop_sizes))
 
 print('Could not find cropped image for the following subjects: ', failed_crop_subjects)
 print('Found shape mismatch in images and GTs for the following subjects: ', shape_mismatch_subjects)