|
a |
|
b/app.py |
|
|
1 |
""" |
|
|
2 |
============================================================================================= |
|
|
3 |
Project : Chest X-Ray Pathology Detection and Localization using Deep Learning |
|
|
4 |
Author Name : Rammuni Ravidu Suien Silva |
|
|
5 |
UoW No : 16267097 |
|
|
6 |
IIT No : 2016134 |
|
|
7 |
Module : Final Year Project 20/21 |
|
|
8 |
Supervisor : Mr Pumudu Fernando |
|
|
9 |
|
|
|
10 |
Prototype : Web Interface - BackEnd [Draft: .v01] |
|
|
11 |
University of Westminster, UK || IIT Sri Lanka |
|
|
12 |
============================================================================================= |
|
|
13 |
""" |
|
|
14 |
import json |
|
|
15 |
import os |
|
|
16 |
from datetime import datetime |
|
|
17 |
|
|
|
18 |
import numpy as np |
|
|
19 |
# Flask Imports |
|
|
20 |
from flask import Flask, request, render_template |
|
|
21 |
from flask import send_file |
|
|
22 |
from flask_jsglue import JSGlue |
|
|
23 |
# Tensorflow Keras imports |
|
|
24 |
from tensorflow.keras.models import load_model |
|
|
25 |
# For secure src links |
|
|
26 |
from werkzeug.utils import secure_filename |
|
|
27 |
|
|
|
28 |
# System Library import |
|
|
29 |
from lab_cxr_scripts.lab_cxr import CXRPrediction, CXRLocalization |
|
|
30 |
|
|
|
31 |
# Model 0, 2 :- xray_labels_set[0] || Model 1 :- xray_labels_set[1] |
|
|
32 |
xray_labels_set = [["Enlarged Cardiomediastinum", "Cardiomegaly", "Lung Lesion", "Lung Opacity", "Edema", |
|
|
33 |
"Consolidation", "Pneumonia", "Atelectasis", "Pneumothorax", "Pleural Effusion", |
|
|
34 |
"Pleural Other", "Fracture", "Support Devices"], |
|
|
35 |
["Nodule", "Cardiomegaly", "Emphysema", "Fibrosis", "Edema", "Consolidation", "Pneumonia", |
|
|
36 |
"Atelectasis", "Pneumothorax", "Pleural Effusion", "Mass", "Infiltration", "Hernia", |
|
|
37 |
"Plueral Thickening"]] |
|
|
38 |
|
|
|
39 |
# Labels for classification tasks |
|
|
40 |
xray_labels = xray_labels_set[0] |
|
|
41 |
# Dependency pip install pyopenssl |
|
|
42 |
# Flask Configs |
|
|
43 |
app = Flask(__name__) |
|
|
44 |
app.config['MAX_CONTENT_LENGTH'] = 20 * 1024 * 1024 # Request data limited to 20MB |
|
|
45 |
jsglue = JSGlue(app) |
|
|
46 |
|
|
|
47 |
# TODO: USER GUIDE |
|
|
48 |
|
|
|
49 |
# model load |
|
|
50 |
models = [[ |
|
|
51 |
load_model('models/MIMIC/PAR-64-MODEL-MIMIC-FINAL-2.h5', |
|
|
52 |
custom_objects={'weighted_loss': CXRPrediction.get_weighted_loss(1, 1)}), |
|
|
53 |
load_model('models/MIMIC/PAR-128-MODEL-MIMIC-FINAL-2.h5', |
|
|
54 |
custom_objects={'weighted_loss': CXRPrediction.get_weighted_loss(1, 1)}) |
|
|
55 |
], [ |
|
|
56 |
load_model('models/NIH/PAR-64-MODEL-FINAL-NIH-2.h5', |
|
|
57 |
custom_objects={'weighted_loss': CXRPrediction.get_weighted_loss(1, 1)}), |
|
|
58 |
load_model('models/NIH/PAR-128-MODEL-FINAL-NIH-2.h5', |
|
|
59 |
custom_objects={'weighted_loss': CXRPrediction.get_weighted_loss(1, 1)}) |
|
|
60 |
]] |
|
|
61 |
model = models[0] |
|
|
62 |
cur_cxr_hash = 'none' |
|
|
63 |
|
|
|
64 |
""" |
|
|
65 |
================================================================================================================== |
|
|
66 |
Web request functions |
|
|
67 |
================================================================================================================== |
|
|
68 |
""" |
|
|
69 |
|
|
|
70 |
|
|
|
71 |
# Web page startup |
|
|
72 |
@app.route('/') |
|
|
73 |
def start_web(): |
|
|
74 |
return render_template("index.html") |
|
|
75 |
|
|
|
76 |
|
|
|
77 |
# CXR Image upload API |
|
|
78 |
@app.route('/predict/<int:model_id>', methods=['GET', 'POST']) |
|
|
79 |
def upload(model_id): |
|
|
80 |
if request.method == 'POST': |
|
|
81 |
print("Model ID", model_id) |
|
|
82 |
|
|
|
83 |
# Selecting Model and labels set |
|
|
84 |
global model, xray_labels |
|
|
85 |
model = models[model_id % len(models)] |
|
|
86 |
xray_labels = xray_labels_set[model_id % len(xray_labels_set)] |
|
|
87 |
|
|
|
88 |
global cur_cxr_hash |
|
|
89 |
preds = [] |
|
|
90 |
file_count = len(request.files) |
|
|
91 |
if file_count > 8: |
|
|
92 |
return |
|
|
93 |
for file_num in range(file_count): |
|
|
94 |
# Getting image file from post request through the Web |
|
|
95 |
cxr_img_file = request.files['file_' + str(file_num)] |
|
|
96 |
# Generating Hash of the image file |
|
|
97 |
hashed_filename = CXRPrediction.hash_cxr(cxr_img_file) |
|
|
98 |
print(hashed_filename) |
|
|
99 |
cur_cxr_hash = hashed_filename |
|
|
100 |
|
|
|
101 |
# Saving the CXR image to uploads |
|
|
102 |
cxr_img_path = os.path.dirname(__file__) |
|
|
103 |
file_path = os.path.join( |
|
|
104 |
cxr_img_path, 'uploads', secure_filename(hashed_filename)) |
|
|
105 |
cxr_img_file.save(file_path) |
|
|
106 |
|
|
|
107 |
# Detection results calculation |
|
|
108 |
preds.append(np.array(CXRPrediction.model_predict(file_path, model)[0]).tolist()) |
|
|
109 |
|
|
|
110 |
# Final results calculation considering the results of all the uploaded images |
|
|
111 |
final_preds = np.round(np.multiply(np.mean(preds, axis=0), 100), 2) |
|
|
112 |
final_preds_max = np.round(np.multiply(np.max(preds, axis=0), 100), 2) |
|
|
113 |
final_preds_min = np.round(np.multiply(np.min(preds, axis=0), 100), 2) |
|
|
114 |
|
|
|
115 |
print(final_preds) |
|
|
116 |
# Creating the detection results dictionary/ JSON |
|
|
117 |
predictions_dict = {} |
|
|
118 |
for i in range(0, len(xray_labels)): |
|
|
119 |
det_rate_str = str(final_preds[i]) + "% (" + str(final_preds_max[i]) + "% - " + str( |
|
|
120 |
final_preds_min[i]) + "%)" |
|
|
121 |
predictions_dict[xray_labels[i]] = det_rate_str |
|
|
122 |
|
|
|
123 |
# Creating detection result JSON to be sent |
|
|
124 |
json_predictions = json.dumps(predictions_dict, indent=4) |
|
|
125 |
|
|
|
126 |
result = json_predictions |
|
|
127 |
return result |
|
|
128 |
return None |
|
|
129 |
|
|
|
130 |
|
|
|
131 |
@app.route('/localize') |
|
|
132 |
def localization(): # Localization API |
|
|
133 |
global cur_cxr_hash |
|
|
134 |
start = datetime.now() |
|
|
135 |
filepath = 'localizations/' + cur_cxr_hash.split('.')[0] |
|
|
136 |
|
|
|
137 |
if os.path.exists(filepath): |
|
|
138 |
file_count = len([name for name in os.listdir(filepath) if os.path.isfile(os.path.join(filepath, name))]) |
|
|
139 |
if not file_count == len(xray_labels): |
|
|
140 |
# If the localized img is already there no need to re-process |
|
|
141 |
CXRLocalization.create_cxr_localization_heatmap(cur_cxr_hash, model[len(model) - 1], xray_labels) |
|
|
142 |
else: |
|
|
143 |
# Calling Localization Function |
|
|
144 |
CXRLocalization.create_cxr_localization_heatmap(cur_cxr_hash, model[len(model) - 1], xray_labels) |
|
|
145 |
|
|
|
146 |
print(datetime.now() - start) |
|
|
147 |
return str(len(xray_labels)) # Returning the localized labels |
|
|
148 |
|
|
|
149 |
|
|
|
150 |
# Function for sending the localized CXR image |
|
|
151 |
@app.route('/get_cxr_detect_img/<int:pathology_id>') |
|
|
152 |
def get_cxr_detect_img(pathology_id): |
|
|
153 |
print(pathology_id) |
|
|
154 |
global cur_cxr_hash |
|
|
155 |
localized_image_name = xray_labels[pathology_id] + '-localizedHeatmap-' + cur_cxr_hash |
|
|
156 |
filepath = 'localizations/' + cur_cxr_hash.split('.')[0] + '/' |
|
|
157 |
return send_file(filepath + localized_image_name, mimetype='image/jpg') |
|
|
158 |
|
|
|
159 |
|
|
|
160 |
# Function for getting symptoms |
|
|
161 |
@app.route('/get_symptoms') |
|
|
162 |
def get_symptoms(): |
|
|
163 |
return send_file('static/files/Symptoms.json', mimetype='application/json') |
|
|
164 |
|
|
|
165 |
|
|
|
166 |
print("Server Running...") |
|
|
167 |
if __name__ == '__main__': |
|
|
168 |
app.run(debug=True) # Debugging |