MASIDevelopment/test.py at main · basicallyAlexOh/MASIDevelopment · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
"""Evaluate model on test set and report results"""

import os
import sys
from tqdm import tqdm
import torch
import SimpleITK as sitk
import gc
import psutil
from monai.inferers import sliding_window_inference
from monai.data import decollate_batch
from monai.transforms import (
    Compose,
    AsDiscrete,
    AddChannel,
    EnsureType,
    EnsureTyped,
    AsDiscreted,
    Spacingd,
    Resized,
    Orientationd,
)
from pandas import DataFrame
import numpy as np
import nibabel as nib
from postprocess import get_largest_cc, lungmask_filling, nearest_label_filling
sys.path.append("/home/local/VANDERBILT/litz/github/MASILab/thoraxtools/func")
import vis.paral_clip_overlay_mask as overlay

def test(config,
         config_id,
         device,
         model,
         model_path,
         test_metric,
         test_loader,
         metrics_path,
         seg_dir,
         clip_dir):

    model.load_state_dict(torch.load(model_path))
    model.eval()
    test_metric.reset()

    # measure_transforms = Compose([EnsureType(), AddChannel(), AsDiscrete(to_onehot=6)])
    image_paths = []
    # reader = sitk.ImageFileReader()
    # print(psutil.Process(os.getpid()).memory_info().rss / 1024 ** 2)
    with torch.no_grad():
        for test_data in tqdm(test_loader):
            test_inputs, image_path = (
                test_data["image"].to(device),
                test_data["image_path"][0],
            )
            # make transforms from image metadata
            raw_nii = nib.load(image_path)
            axcodes = nib.orientations.aff2axcodes(raw_nii.affine)
            axcodes = ''.join(axcodes)
            pixdim = raw_nii.header.get_zooms()
            spatial_size = raw_nii.shape
            post_pred_transforms = Compose([
                EnsureTyped(keys=["label", "pred"]),
                AsDiscreted(keys="pred", argmax=True, to_onehot=6),
                AsDiscreted(keys="label", to_onehot=6),
                Orientationd(keys=["pred"], axcodes=axcodes),
                Spacingd(keys=["pred"], pixdim=pixdim, mode="nearest"),
                Resized(keys=["pred"], spatial_size=spatial_size, mode="nearest"),
            ])

            image_paths.append(image_path)
            test_data["pred"] = sliding_window_inference(test_inputs, config["crop_shape"], 4, model)
            test_data["pred"] = test_data["pred"].detach().cpu() # image space transform is mem intensive
            test_data = [post_pred_transforms(i) for i in decollate_batch(test_data)]
            test_data = test_data[0]

            # postprocces with CC and nearest label filling
            pred = torch.argmax(test_data["pred"], dim=0)
            # pred = lungmask_filling(get_largest_cc(pred), image_path)
            pred = nearest_label_filling(pred, get_largest_cc(pred))
            measure_transforms = Compose([EnsureType(), AddChannel(), AsDiscrete(to_onehot=len(np.unique(pred)))])
            test_data["pred"] = measure_transforms(pred)

            # compute dice
            test_metric(y_pred=[test_data["pred"]], y=[test_data["label"]])
            # final label map
            label_map = torch.argmax(test_data["pred"], dim=0)

            # dice of this example's RML
            # rml_dice = test_metric.aggregate()[-1][3]

            # pred_img_space = invert_transforms(test_data)["pred"]
            # pred_img_space = torch.argmax(pred_img_space, dim=0)
            #
            # if seg_dir and rml_dice < 0.7:
            # # # if seg_dir:
            #     print(psutil.Process(os.getpid()).memory_info().rss / 1024 ** 2)
            #     reader.SetFileName(image_path)
            #     reader.ReadImageInformation()
            #     seg = np.transpose(pred_img_space, (2, 1, 0))
            #     seg = sitk.GetImageFromArray(seg)
            #     seg.SetOrigin(reader.GetOrigin())
            #     seg.SetSpacing(reader.GetSpacing())
            #     seg.SetDirection(reader.GetDirection())
            #     sitk.WriteImage(seg, os.path.join(seg_dir, os.path.basename(image_path)))
            #     print(psutil.Process(os.getpid()).memory_info().rss / 1024 ** 2)
            # if clip_dir and rml_dice < 0.7:
            # # if clip_dir:
            #     # raw_sitk = sitk.ReadImage(image_path)
            #     # raw = sitk.GetArrayFromImage(raw_sitk)
            #     # raw = test_data["image"]
            #     raw = test_data["image"].detach().cpu().numpy()
            #     raw = raw[0,:,:,:]
            #     # fname = os.path.basename(image_path)[:-4] if config["dataset"]=="luna16" else os.path.basename(image_path).split(".")[0]
            #     fname = os.path.basename(image_path)[:-7]
            #     overlay.multiple_clip_overlay_with_mask_from_npy(raw, postprocess, os.path.join(clip_dir, f"pred_{fname}_coronal.png"), clip_plane='coronal', img_vrange=(0, 1))
            #     overlay.multiple_clip_overlay_with_mask_from_npy(raw, label, os.path.join(clip_dir, f"label_{fname}_coronal.png"), clip_plane='coronal', img_vrange=(0, 1))
            #     overlay.multiple_clip_overlay_with_mask_from_npy(raw, postprocess,
            #                                                      os.path.join(clip_dir, f"pred_{fname}_axial.png"),
            #                                                      clip_plane='axial', img_vrange=(0, 1))
            #     overlay.multiple_clip_overlay_with_mask_from_npy(raw, label,
            #                                                  os.path.join(clip_dir, f"label_{fname}_axial.png"),
            #                                                  clip_plane='axial', img_vrange=(0, 1))
                # overlay.multiple_clip_overlay_with_mask_from_npy(raw, postprocess, os.path.join(clip_dir, f"pred_{fname}_sagittal.png"), clip_plane='sagittal', img_vrange=(0,1))
                # overlay.multiple_clip_overlay_with_mask_from_npy(raw, postprocess, os.path.join(clip_dir, f"pred_{fname}_axial.png"), clip_plane='axial', img_vrange=(0,1))
                # print(psutil.Process(os.getpid()).memory_info().rss / 1024 ** 2)
            gc.collect()

        # Total dice over test set
        test_dices = test_metric.aggregate()

        # Record metrics and compute mean over test set
        class_means = torch.mean(test_dices, dim=0)
        mean = torch.mean(test_dices)

        # store in dataframe with image path
        test_dices_df = DataFrame(test_dices.detach().cpu().numpy())
        test_dices_df["input_path"] = image_paths
        test_dices_df.to_csv(metrics_path)

    # Log best dice
    # print(f"All scores: {test_dices_df}")
    print(f"Average class scores: {class_means}")
    print(f"Average score overall: {mean}")