Chris Oswald
commited on
Commit
·
508d191
1
Parent(s):
01d5f4f
added image standardization
Browse files
SPIDER.py
CHANGED
|
@@ -24,6 +24,7 @@ import numpy as np
|
|
| 24 |
import pandas as pd
|
| 25 |
|
| 26 |
import datasets
|
|
|
|
| 27 |
import SimpleITK as sitk
|
| 28 |
|
| 29 |
# Define functions
|
|
@@ -36,12 +37,20 @@ def import_csv_data(filepath: str) -> List[Dict[str, str]]:
|
|
| 36 |
results.append(line)
|
| 37 |
return results
|
| 38 |
|
| 39 |
-
def standardize_3D_image(
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
if image.shape[0] < image.shape[2]:
|
| 42 |
image = np.transpose(image, axes=[1, 2, 0])
|
|
|
|
|
|
|
| 43 |
return image
|
| 44 |
|
|
|
|
| 45 |
# Define constants
|
| 46 |
N_PATIENTS = 218
|
| 47 |
MIN_IVD = 0
|
|
@@ -91,11 +100,11 @@ class CustomBuilderConfig(datasets.BuilderConfig):
|
|
| 91 |
data_files: Optional[Union[str, Sequence, Mapping]] = None,
|
| 92 |
description: Optional[str] = None,
|
| 93 |
scan_types: List[str] = DEFAULT_SCAN_TYPES,
|
| 94 |
-
|
| 95 |
):
|
| 96 |
super().__init__(name, version, data_dir, data_files, description)
|
| 97 |
self.scan_types = scan_types
|
| 98 |
-
self.
|
| 99 |
|
| 100 |
|
| 101 |
class SPIDER(datasets.GeneratorBasedBuilder):
|
|
@@ -111,28 +120,28 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 111 |
# version=VERSION,
|
| 112 |
# description="Use images of all scan types (t1, t2, t2 SPACE)",
|
| 113 |
# scan_types=['t1', 't2', 't2_SPACE'],
|
| 114 |
-
#
|
| 115 |
# ),
|
| 116 |
# CustomBuilderConfig(
|
| 117 |
# name="t1_scan_types",
|
| 118 |
# version=VERSION,
|
| 119 |
# description="Use images of t1 scan types only",
|
| 120 |
# scan_types=['t1'],
|
| 121 |
-
#
|
| 122 |
# ),
|
| 123 |
# CustomBuilderConfig(
|
| 124 |
# name="t2_scan_types",
|
| 125 |
# version=VERSION,
|
| 126 |
# description="Use images of t2 scan types only",
|
| 127 |
# scan_types=['t2'],
|
| 128 |
-
#
|
| 129 |
# ),
|
| 130 |
# CustomBuilderConfig(
|
| 131 |
# name="t2_SPACE_scan_types",
|
| 132 |
# version=VERSION,
|
| 133 |
# description="Use images of t2 SPACE scan types only",
|
| 134 |
# scan_types=['t2_SPACE'],
|
| 135 |
-
#
|
| 136 |
# ),
|
| 137 |
# ]
|
| 138 |
|
|
@@ -142,12 +151,12 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 142 |
self,
|
| 143 |
*args,
|
| 144 |
scan_types: List[str] = DEFAULT_SCAN_TYPES,
|
| 145 |
-
|
| 146 |
**kwargs,
|
| 147 |
):
|
| 148 |
super().__init__(*args, **kwargs)
|
| 149 |
self.scan_types = scan_types
|
| 150 |
-
self.
|
| 151 |
|
| 152 |
def _info(self):
|
| 153 |
"""
|
|
@@ -158,7 +167,7 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 158 |
"patient_id": datasets.Value("string"),
|
| 159 |
"scan_type": datasets.Value("string"),
|
| 160 |
# "raw_image": datasets.Image(),
|
| 161 |
-
"numeric_array": datasets.Array3D(shape=self.
|
| 162 |
"metadata": {
|
| 163 |
"num_vertebrae": datasets.Value(dtype="string"),
|
| 164 |
"num_discs": datasets.Value(dtype="string"),
|
|
@@ -246,7 +255,7 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 246 |
"paths_dict": paths_dict,
|
| 247 |
"split": "train",
|
| 248 |
"scan_types": self.scan_types,
|
| 249 |
-
"
|
| 250 |
},
|
| 251 |
),
|
| 252 |
datasets.SplitGenerator(
|
|
@@ -255,7 +264,7 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 255 |
"paths_dict": paths_dict,
|
| 256 |
"split": "validate",
|
| 257 |
"scan_types": self.scan_types,
|
| 258 |
-
"
|
| 259 |
},
|
| 260 |
),
|
| 261 |
datasets.SplitGenerator(
|
|
@@ -264,7 +273,7 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 264 |
"paths_dict": paths_dict,
|
| 265 |
"split": "test",
|
| 266 |
"scan_types": self.scan_types,
|
| 267 |
-
"
|
| 268 |
},
|
| 269 |
),
|
| 270 |
]
|
|
@@ -274,7 +283,7 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 274 |
paths_dict: Dict[str, str],
|
| 275 |
split: str,
|
| 276 |
scan_types: List[str],
|
| 277 |
-
|
| 278 |
validate_share: float = 0.3,
|
| 279 |
test_share: float = 0.2,
|
| 280 |
raw_image: bool = True,
|
|
@@ -475,9 +484,11 @@ class SPIDER(datasets.GeneratorBasedBuilder):
|
|
| 475 |
image_path = os.path.join(paths_dict['images'], 'images', example)
|
| 476 |
image = sitk.ReadImage(image_path)
|
| 477 |
|
| 478 |
-
# Convert .mha image to numeric array
|
| 479 |
-
image_array =
|
| 480 |
-
|
|
|
|
|
|
|
| 481 |
#TODO: load mask file and numeric array
|
| 482 |
|
| 483 |
# Extract overview data corresponding to image
|
|
|
|
| 24 |
import pandas as pd
|
| 25 |
|
| 26 |
import datasets
|
| 27 |
+
import skimage
|
| 28 |
import SimpleITK as sitk
|
| 29 |
|
| 30 |
# Define functions
|
|
|
|
| 37 |
results.append(line)
|
| 38 |
return results
|
| 39 |
|
| 40 |
+
def standardize_3D_image(
|
| 41 |
+
image: np.ndarray,
|
| 42 |
+
resize_shape: Tuple[int, int, int]
|
| 43 |
+
) -> np.ndarray:
|
| 44 |
+
"""Aligns dimensions of image to be (height, width, channels) and resizes
|
| 45 |
+
images to values specified in resize_shape."""
|
| 46 |
+
# Align height, width, channel dims
|
| 47 |
if image.shape[0] < image.shape[2]:
|
| 48 |
image = np.transpose(image, axes=[1, 2, 0])
|
| 49 |
+
# Resize image
|
| 50 |
+
image = skimage.transform.resize(image, resize_shape)
|
| 51 |
return image
|
| 52 |
|
| 53 |
+
|
| 54 |
# Define constants
|
| 55 |
N_PATIENTS = 218
|
| 56 |
MIN_IVD = 0
|
|
|
|
| 100 |
data_files: Optional[Union[str, Sequence, Mapping]] = None,
|
| 101 |
description: Optional[str] = None,
|
| 102 |
scan_types: List[str] = DEFAULT_SCAN_TYPES,
|
| 103 |
+
resize_shape: Tuple[int, int, int] = DEFAULT_RESIZE,
|
| 104 |
):
|
| 105 |
super().__init__(name, version, data_dir, data_files, description)
|
| 106 |
self.scan_types = scan_types
|
| 107 |
+
self.resize_shape = resize_shape
|
| 108 |
|
| 109 |
|
| 110 |
class SPIDER(datasets.GeneratorBasedBuilder):
|
|
|
|
| 120 |
# version=VERSION,
|
| 121 |
# description="Use images of all scan types (t1, t2, t2 SPACE)",
|
| 122 |
# scan_types=['t1', 't2', 't2_SPACE'],
|
| 123 |
+
# resize_shape=DEFAULT_RESIZE,
|
| 124 |
# ),
|
| 125 |
# CustomBuilderConfig(
|
| 126 |
# name="t1_scan_types",
|
| 127 |
# version=VERSION,
|
| 128 |
# description="Use images of t1 scan types only",
|
| 129 |
# scan_types=['t1'],
|
| 130 |
+
# resize_shape=DEFAULT_RESIZE,
|
| 131 |
# ),
|
| 132 |
# CustomBuilderConfig(
|
| 133 |
# name="t2_scan_types",
|
| 134 |
# version=VERSION,
|
| 135 |
# description="Use images of t2 scan types only",
|
| 136 |
# scan_types=['t2'],
|
| 137 |
+
# resize_shape=DEFAULT_RESIZE,
|
| 138 |
# ),
|
| 139 |
# CustomBuilderConfig(
|
| 140 |
# name="t2_SPACE_scan_types",
|
| 141 |
# version=VERSION,
|
| 142 |
# description="Use images of t2 SPACE scan types only",
|
| 143 |
# scan_types=['t2_SPACE'],
|
| 144 |
+
# resize_shape=DEFAULT_RESIZE,
|
| 145 |
# ),
|
| 146 |
# ]
|
| 147 |
|
|
|
|
| 151 |
self,
|
| 152 |
*args,
|
| 153 |
scan_types: List[str] = DEFAULT_SCAN_TYPES,
|
| 154 |
+
resize_shape: Tuple[int, int, int] = DEFAULT_RESIZE,
|
| 155 |
**kwargs,
|
| 156 |
):
|
| 157 |
super().__init__(*args, **kwargs)
|
| 158 |
self.scan_types = scan_types
|
| 159 |
+
self.resize_shape = resize_shape
|
| 160 |
|
| 161 |
def _info(self):
|
| 162 |
"""
|
|
|
|
| 167 |
"patient_id": datasets.Value("string"),
|
| 168 |
"scan_type": datasets.Value("string"),
|
| 169 |
# "raw_image": datasets.Image(),
|
| 170 |
+
"numeric_array": datasets.Array3D(shape=self.resize_shape, dtype='int16'),
|
| 171 |
"metadata": {
|
| 172 |
"num_vertebrae": datasets.Value(dtype="string"),
|
| 173 |
"num_discs": datasets.Value(dtype="string"),
|
|
|
|
| 255 |
"paths_dict": paths_dict,
|
| 256 |
"split": "train",
|
| 257 |
"scan_types": self.scan_types,
|
| 258 |
+
"resize_shape": self.resize_shape,
|
| 259 |
},
|
| 260 |
),
|
| 261 |
datasets.SplitGenerator(
|
|
|
|
| 264 |
"paths_dict": paths_dict,
|
| 265 |
"split": "validate",
|
| 266 |
"scan_types": self.scan_types,
|
| 267 |
+
"resize_shape": self.resize_shape,
|
| 268 |
},
|
| 269 |
),
|
| 270 |
datasets.SplitGenerator(
|
|
|
|
| 273 |
"paths_dict": paths_dict,
|
| 274 |
"split": "test",
|
| 275 |
"scan_types": self.scan_types,
|
| 276 |
+
"resize_shape": self.resize_shape,
|
| 277 |
},
|
| 278 |
),
|
| 279 |
]
|
|
|
|
| 283 |
paths_dict: Dict[str, str],
|
| 284 |
split: str,
|
| 285 |
scan_types: List[str],
|
| 286 |
+
resize_shape: Tuple[int, int, int],
|
| 287 |
validate_share: float = 0.3,
|
| 288 |
test_share: float = 0.2,
|
| 289 |
raw_image: bool = True,
|
|
|
|
| 484 |
image_path = os.path.join(paths_dict['images'], 'images', example)
|
| 485 |
image = sitk.ReadImage(image_path)
|
| 486 |
|
| 487 |
+
# Convert .mha image to standardized numeric array
|
| 488 |
+
image_array = standardize_3D_image(
|
| 489 |
+
sitk.GetArrayFromImage(image), resize_shape
|
| 490 |
+
)
|
| 491 |
+
|
| 492 |
#TODO: load mask file and numeric array
|
| 493 |
|
| 494 |
# Extract overview data corresponding to image
|