MRSegmentator / utils.py
DiGuaQiu's picture
Create utils.py
aa2b9a2 verified
raw
history blame
6.44 kB
# Code copied and modified from https://huggingface.co/spaces/BAAI/SegVol/blob/main/utils.py
from pathlib import Path
import matplotlib as mpl
import matplotlib.pyplot as plt
import nibabel as nib
import numpy as np
import torch
from monai.transforms import LoadImage
from mrsegmentator import inference
from mrsegmentator.utils import add_postfix
from PIL import Image, ImageColor, ImageDraw, ImageEnhance
from scipy import ndimage
from monai.transforms import LoadImage, Orientation, Spacing
import SimpleITK as sitk
import streamlit as st
initial_rectangle = {
"version": "4.4.0",
"objects": [
{
"type": "rect",
"version": "4.4.0",
"originX": "left",
"originY": "top",
"left": 50,
"top": 50,
"width": 100,
"height": 100,
"fill": "rgba(255, 165, 0, 0.3)",
"stroke": "#2909F1",
"strokeWidth": 3,
"strokeDashArray": None,
"strokeLineCap": "butt",
"strokeDashOffset": 0,
"strokeLineJoin": "miter",
"strokeUniform": True,
"strokeMiterLimit": 4,
"scaleX": 1,
"scaleY": 1,
"angle": 0,
"flipX": False,
"flipY": False,
"opacity": 1,
"shadow": None,
"visible": True,
"backgroundColor": "",
"fillRule": "nonzero",
"paintFirst": "fill",
"globalCompositeOperation": "source-over",
"skewX": 0,
"skewY": 0,
"rx": 0,
"ry": 0,
}
],
}
def run(tmpdirname):
if st.session_state.option is not None:
image = Path(__file__).parent / str(st.session_state.option)
inference.infer([image], tmpdirname, [0], split_level=1)
seg_name = add_postfix(image.name, "seg")
preds_path = tmpdirname + "/" + seg_name
st.session_state.preds_3D = read_image(preds_path)
st.session_state.preds_3D_ori = sitk.ReadImage(preds_path)
def reflect_box_into_model(box_3d):
z1, y1, x1, z2, y2, x2 = box_3d
x1_prompt = int(x1 * 256.0 / 325.0)
y1_prompt = int(y1 * 256.0 / 325.0)
z1_prompt = int(z1 * 32.0 / 325.0)
x2_prompt = int(x2 * 256.0 / 325.0)
y2_prompt = int(y2 * 256.0 / 325.0)
z2_prompt = int(z2 * 32.0 / 325.0)
return torch.tensor(
np.array([z1_prompt, y1_prompt, x1_prompt, z2_prompt, y2_prompt, x2_prompt])
)
def reflect_json_data_to_3D_box(json_data, view):
if view == "xy":
st.session_state.rectangle_3Dbox[1] = json_data["objects"][0]["top"]
st.session_state.rectangle_3Dbox[2] = json_data["objects"][0]["left"]
st.session_state.rectangle_3Dbox[4] = (
json_data["objects"][0]["top"]
+ json_data["objects"][0]["height"] * json_data["objects"][0]["scaleY"]
)
st.session_state.rectangle_3Dbox[5] = (
json_data["objects"][0]["left"]
+ json_data["objects"][0]["width"] * json_data["objects"][0]["scaleX"]
)
print(st.session_state.rectangle_3Dbox)
def make_fig(image, preds, px_range = (10, 400), transparency=0.5):
fig, ax = plt.subplots(1, 1, figsize=(4,4))
image_slice = image.clip(*px_range)
ax.imshow(
image_slice,
cmap="Greys_r",
vmin=px_range[0],
vmax=px_range[1],
)
if preds is not None:
image_slice = np.array(preds)
alpha = np.zeros(image_slice.shape)
alpha[image_slice > 0.1] = transparency
ax.imshow(
image_slice,
cmap="jet",
alpha=alpha,
vmin=0,
vmax=40,
)
# plot edges
edge_slice = np.zeros(image_slice.shape, dtype=int)
for i in np.unique(image_slice):
_slice = image_slice.copy()
_slice[_slice != i] = 0
edges = ndimage.laplace(_slice)
edge_slice[edges != 0] = i
cmap = mpl.cm.jet(np.linspace(0, 1, int(preds.max())))
cmap -= 0.4
cmap = cmap.clip(0, 1)
cmap = mpl.colors.ListedColormap(cmap)
alpha = np.zeros(edge_slice.shape)
alpha[edge_slice > 0.01] = 0.9
ax.imshow(
edge_slice,
alpha=alpha,
cmap=cmap,
vmin=0,
vmax=40,
)
plt.axis("off")
ax.set_xticks([])
ax.set_yticks([])
fig.canvas.draw()
# transform to image
return Image.frombytes("RGB", fig.canvas.get_width_height(), fig.canvas.tostring_rgb())
#######################################
def make_isotropic(image, interpolator = sitk.sitkLinear, spacing = None):
'''
Many file formats (e.g. jpg, png,...) expect the pixels to be isotropic, same
spacing for all axes. Saving non-isotropic data in these formats will result in
distorted images. This function makes an image isotropic via resampling, if needed.
Args:
image (SimpleITK.Image): Input image.
interpolator: By default the function uses a linear interpolator. For
label images one should use the sitkNearestNeighbor interpolator
so as not to introduce non-existant labels.
spacing (float): Desired spacing. If none given then use the smallest spacing from
the original image.
Returns:
SimpleITK.Image with isotropic spacing which occupies the same region in space as
the input image.
'''
original_spacing = image.GetSpacing()
# Image is already isotropic, just return a copy.
if all(spc == original_spacing[0] for spc in original_spacing):
return sitk.Image(image)
# Make image isotropic via resampling.
original_size = image.GetSize()
if spacing is None:
spacing = min(original_spacing)
new_spacing = [spacing]*image.GetDimension()
new_size = [int(round(osz*ospc/spacing)) for osz, ospc in zip(original_size, original_spacing)]
return sitk.Resample(image, new_size, sitk.Transform(), interpolator,
image.GetOrigin(), new_spacing, image.GetDirection(), 0, # default pixel value
image.GetPixelID())
def read_image(path):
img = sitk.ReadImage(path)
img = sitk.DICOMOrient(img, "LPS")
img = make_isotropic(img)
img = sitk.GetArrayFromImage(img)
return img