MSHGC/utils.py at main · iMoonLab/MSHGC · 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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
import json
import os.path as osp
from sklearn.model_selection import KFold
import glob
import numpy as np
import torch
import lifelines.utils.concordance as LUC
import random
import pickle


GENDER = {'male':0,'female':1}
def get_WSI_sample_list(WSI_info_list,centers,patch_ft_dir,WSI_patch_coor_dir,multi_label=False):
    with open(WSI_info_list, 'r') as fp:
        lbls = json.load(fp)
    if WSI_patch_coor_dir is not None:
        all_coor_list = []
        if isinstance(centers,list):
            for center in centers:
                all_coor_list.extend(glob.glob(osp.join(WSI_patch_coor_dir.format(center), '*_coors.pkl')))
        else:
            all_coor_list.extend(glob.glob(osp.join(WSI_patch_coor_dir.format(centers), '*_coors.pkl')))
        coor_dict = {}
        def get_id(_dir):
            tmp_dir = _dir.split('/')[-1][:-14]
            return tmp_dir
        for _dir in all_coor_list:
            id = get_id(_dir)
            coor_dict[id] = _dir

    if patch_ft_dir is not None:
        all_ft_list = []
        if isinstance(centers,list):
            for center in centers:
                all_ft_list.extend(glob.glob(osp.join(patch_ft_dir.format(center), '*_fts.npy')))
        else:
            all_ft_list.extend(glob.glob(osp.join(patch_ft_dir.format(centers), '*_fts.npy')))
        ft_dict = {}
        def get_id(_dir):
            tmp_dir = _dir.split('/')[-1][:-12]
            return tmp_dir
        for _dir in all_ft_list:
            id = get_id(_dir)
            ft_dict[id] = _dir


    all_dict = {}
    survival_time_max = 0
    survival_time_min = None

    for patient in lbls.keys():
        all_dict[patient] = {}
        time = int(lbls[patient]['OS-time'])
        all_dict[patient]['survival_time'] = time

        all_dict[patient]['WSIs'] = {}
        for file in lbls[patient]['WSI_files']:
            wsi_id = file[:-4]
            all_dict[patient]['WSIs'][wsi_id] = {}
            if str(wsi_id) in coor_dict.keys():
                all_dict[patient]['WSIs'][wsi_id]['patch_coors'] = coor_dict[str(wsi_id)]
            else:
                print(wsi_id)
            if str(wsi_id) in ft_dict.keys():
                all_dict[patient]['WSIs'][wsi_id]['ft_dir'] = ft_dict[str(wsi_id)]
            else:
                print(wsi_id)

        all_dict[patient]['status'] = int(lbls[patient]['OS'])

        if multi_label:
            all_dict[patient]['T stage'] = int(lbls[patient]['T_stage'])
            all_dict[patient]['N stage'] = int(lbls[patient]['N_stage'])
            all_dict[patient]['M stage'] = int(lbls[patient]['M_stage'])
            all_dict[patient]['TNM stage'] = int(lbls[patient]['TNM_stage'])

        survival_time_max = survival_time_max \
            if survival_time_max > time else time
        if survival_time_min is None:
            survival_time_min = time
        else:
            survival_time_min = survival_time_min \
                if survival_time_min < time else time


    return all_dict, survival_time_max, survival_time_min

def get_n_fold_data_list(data_dict,n_fold,random_seed):
    censored_keys = []
    uncensored_keys = []
    for key in data_dict.keys():
        if data_dict[key]['status'] == 1:
            uncensored_keys.append(key)
        else:
            censored_keys.append(key)
    print("censored length {}".format(len(censored_keys)))
    print("uncensored length {}".format(len(uncensored_keys)))

    n_fold_uncensored_train_list = []
    n_fold_uncensored_val_list = []
    n_fold_censored_train_list = []
    n_fold_censored_val_list = []
    n_fold_train_list = []
    n_fold_val_list = []
    kf = KFold(n_splits=n_fold, shuffle=True, random_state=random_seed) #random_seed
    for train_idx, val_idx in kf.split(uncensored_keys):
        train_keys = [uncensored_keys[i] for i in train_idx]
        val_keys = [uncensored_keys[i] for i in val_idx]

        train_data_dict = {key: data_dict[key] for key in train_keys}
        val_data_dict = {key: data_dict[key] for key in val_keys}
        n_fold_uncensored_train_list.append(train_data_dict)
        n_fold_uncensored_val_list.append(val_data_dict)

    for train_idx, val_idx in kf.split(censored_keys):
        train_keys = [censored_keys[i] for i in train_idx]
        val_keys = [censored_keys[i] for i in val_idx]

        train_data_dict = {key: data_dict[key] for key in train_keys}
        val_data_dict = {key: data_dict[key] for key in val_keys}
        n_fold_censored_train_list.append(train_data_dict)
        n_fold_censored_val_list.append(val_data_dict)

    for i in range(n_fold):
        n_fold_train_list.append(dict(n_fold_censored_train_list[i],**n_fold_uncensored_train_list[i]))
        n_fold_val_list.append(dict(n_fold_censored_val_list[i],**n_fold_uncensored_val_list[i]))

    return n_fold_train_list, n_fold_val_list

def sort_survival_time(gt_survival_time,pre_risk,censore, output_fts=None,patch_ft=None,coors=None):
    ix = torch.argsort(gt_survival_time, dim= 0, descending=True)#
    gt_survival_time = gt_survival_time[ix]
    pre_risk = pre_risk[ix]
    censore = censore[ix]
    # output_fts = output_fts[ix]
    if patch_ft is not None:
        patch_ft = patch_ft[ix]
        coors = coors[ix]
        return gt_survival_time,pre_risk,censore,output_fts,patch_ft,coors
    return gt_survival_time,pre_risk,censore#,output_fts

def accuracytest(survivals, risk, censors):
    survlist = []
    risklist = []
    censorlist = []

    for riskval in risk:
        # riskval = -riskval
        risklist.append(riskval.cpu().detach().item())

    for censorval in censors:
        censorlist.append(censorval.cpu().detach().item())

    for surval in survivals:
        # surval = -surval
        survlist.append(surval.cpu().detach().item())

    C_value = LUC.concordance_index(survlist, risklist, censorlist)

    return C_value