├── LuncCancerTrain.py
├── NeuralNetwork.py
├── PredictCancer.py
├── README.md
└── dataset_create.py


/LuncCancerTrain.py:
--------------------------------------------------------------------------------
  1 | import cv2
  2 | from scipy import ndimage
  3 | import numpy as np
  4 | import os.path
  5 | from math import sqrt 
  6 | 
  7 | def preprocess(img):
  8 | 	img_filted=cv2.medianBlur(img,5)#5 is mask size
  9 | 	thresh,img_thresh=cv2.threshold(img_filted,127,255,cv2.THRESH_BINARY)
 10 | 	kernel=np.ones((7,7),np.uint8)
 11 | 	img_opened=cv2.morphologyEx(img_thresh,cv2.MORPH_OPEN,kernel)
 12 | 	img_closed=cv2.morphologyEx(img_opened,cv2.MORPH_CLOSE,kernel)
 13 | 	return img_closed,img_filted,img_thresh,img_opened
 14 | 
 15 | def lung_side_detect(c_index,contours,img):
 16 | 	img_roied_left_lung=img
 17 | 	img_roied_right_lung=img
 18 | 	#left lung check
 19 | 	x,y,w,h=cv2.boundingRect(contours[c_index[0]])
 20 | 	M=cv2.moments(contours[c_index[0]])
 21 | 	cx=int(M['m10']/M['m00'])
 22 | 	cy=int(M['m01']/M['m00'])
 23 | 	#print(cx,cy,int(img.shape[1]/2))
 24 | 	if(cx<int(img.shape[1]/2)):
 25 | 		img_roied_left_lung=img[y:y+h,x:x+w]
 26 | 	else:
 27 | 		img_roied_right_lung=img[y:y+h,x:x+w]
 28 | 	#right lung check
 29 | 	x,y,w,h=cv2.boundingRect(contours[c_index[1]])
 30 | 	M=cv2.moments(contours[c_index[1]])
 31 | 	cx=int(M['m10']/M['m00'])
 32 | 	cy=int(M['m01']/M['m00'])
 33 | 	#print(cx,cy,int(img.shape[1]/2))
 34 | 	if(cx<int(img.shape[1]/2)):
 35 | 		img_roied_left_lung=img[y:y+h,x:x+w]
 36 | 	else:
 37 | 		img_roied_right_lung=img[y:y+h,x:x+w]
 38 | 
 39 | 	return img_roied_left_lung,img_roied_right_lung
 40 | 
 41 | def seg_cont(img):
 42 | 	img_cont,contours,hierarchy=cv2.findContours(img,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
 43 | 	zipped=zip([i for i in range(len(contours))],[cv2.contourArea(contours[i]) for i in range(len(contours))])
 44 | 	zipped=list(reversed(sorted(zipped,key=lambda x:x[1])))
 45 | 	c_index=[zipped[1][0],zipped[2][0]]
 46 | 	img_roied_left_lung,img_roied_right_lung=lung_side_detect(c_index,contours,img)
 47 | 	return img_roied_left_lung,img_roied_right_lung
 48 | 
 49 | def sub_feature_extract(sub_image):
 50 | 	feature_list = []
 51 | 	height,width = sub_image.shape
 52 | 	total_pix = height * width		
 53 | 	sub_image_1 = sub_image[0:int(height/2),0:width]		
 54 | 	sub_image_2 = sub_image[int(height/2):height,0:width]
 55 | 	sub_image_3 = sub_image[0:height,0:int(width/2)] 
 56 | 	sub_image_4 = sub_image[0:height,int(width/2):width]
 57 | 	unique,counts=np.unique(sub_image_1,return_counts=True)#gives number of white and black pix
 58 | 	feature_list.append(counts[0]/total_pix)#feature1=black_pix_up/tot_pix
 59 | 	unique,counts=np.unique(sub_image_2,return_counts=True)#gives number of white and black pix
 60 | 	feature_list.append(counts[0]/total_pix)#feature2=black_pix_down/tot_pix
 61 | 	unique,counts=np.unique(sub_image_3,return_counts=True)#gives number of white and black pix
 62 | 	feature_list.append(counts[0]/total_pix)#feature3=black_pix_left/tot_pix
 63 | 	unique,counts=np.unique(sub_image_4,return_counts=True)#gives number of white and black pix
 64 | 	feature_list.append(counts[0]/total_pix)#feature4=black_pix_right/tot_pix	
 65 | 	return	feature_list
 66 | 
 67 | def sub_feature_extract_2(sub_image):
 68 | 	feature_list = []
 69 | 	height,width=sub_image.shape
 70 | 	total_pix = height*width
 71 | 	sub_image_1 = sub_image[0:int(height/2),0:int(width/2)]
 72 | 	sub_image_2 = sub_image[0:int(height/2),int(width/2):width]
 73 | 	sub_image_3 = sub_image[int(height/2):height,0:int(width/2)]
 74 | 	sub_image_4 = sub_image[int(height/2):height,int(width/2):width]
 75 | 	unique,counts=np.unique(sub_image_1,return_counts=True)#gives number of white and black pix
 76 | 	feature_list.append(counts[0]/total_pix)#feature1=black_pix_up_left/tot_pix
 77 | 	unique,counts=np.unique(sub_image_2,return_counts=True)#gives number of white and black pix
 78 | 	feature_list.append(counts[0]/total_pix)#feature2=black_pix_up_right/tot_pix
 79 | 	unique,counts=np.unique(sub_image_3,return_counts=True)#gives number of white and black pix
 80 | 	feature_list.append(counts[0]/total_pix)#feature3=black_pix_down_left/tot_pix
 81 | 	unique,counts=np.unique(sub_image_4,return_counts=True)#gives number of white and black pix
 82 | 	feature_list.append(counts[0]/total_pix)#feature4=black_pix_down_right/tot_pix	
 83 | 	return	feature_list
 84 | 
 85 | 	
 86 | def feature_extract(lung_image):
 87 | 	feature_vector=[]
 88 | 	height,width=lung_image.shape#height is shape(0)
 89 | 	#print(shape)
 90 | 	cent_x=int(width/2)
 91 | 	cent_y=int(height/2)
 92 | 	feature_vector.append(height/width)#feature1=height/width
 93 | 	total_pix=height*width
 94 | 	feature_list = sub_feature_extract(lung_image)#feature2,3,4,5 
 95 | 	feature_vector = feature_vector + feature_list
 96 | 	feature_list = sub_feature_extract_2(lung_image)#feature 6,7,8,9
 97 | 	feature_vector = feature_vector + feature_list
 98 | 	feature_list = sub_feature_extract_2(lung_image[0:int(height/2),0:int(width/2)])#feature 10,11,12,13
 99 | 	feature_vector = feature_vector + feature_list
100 | 	feature_list = sub_feature_extract_2(lung_image[0:int(height/2),int(width/2):width])#feature 14,15,16,17
101 | 	feature_vector = feature_vector + feature_list
102 | 	feature_list = sub_feature_extract_2(lung_image[int(height/2):height,0:int(width/2)])#feature 18,19,20,21
103 | 	feature_vector = feature_vector + feature_list
104 | 	feature_list = sub_feature_extract_2(lung_image[int(height/2):height,int(width/2):width])#feature 22,23,24,25
105 | 	feature_vector = feature_vector + feature_list
106 | 	
107 | 	feature_26_list = []	
108 | 	m00 = []
109 | 	m10 = []
110 | 	m01 = []
111 | 	for i in range(height): #y vale
112 | 		for j in range(width): #x value
113 | 			if lung_image[i,j] == 0:
114 | 				feature_26_list.append(sqrt(((j - cent_x)**2) + ((i - cent_y)**2)))
115 | 			m00.append(lung_image[i,j])
116 | 			m10.append(j * lung_image[i,j])
117 | 			m01.append(i * lung_image[i,j])
118 | 	feature_vector.append(sum(feature_26_list)/total_pix) #feature 26
119 | 	m00 = sum(m00)
120 | 	m10 = sum(m10)
121 | 	m01 = sum(m01)
122 | 	x_bar = m10/m00
123 | 	y_bar = m01/m00
124 | 	
125 | 	u = [[[],[],[],[]],[[],[],[],[]],[[],[],[],[]],[[],[],[],[]]]
126 | 	for i in range(height): #y vale
127 | 		for j in range(width): #x value
128 | 			for p in range(4):
129 | 				for q in range(4):
130 | 					u[p][q].append(((j-x_bar)**p)*((i-y_bar)**q)*lung_image[i,j])	 
131 | 	neta = [[0,0,0,0],[0,0,0,0],[0,0,0,0],[0,0,0,0]]
132 | 	for p in range(4):
133 | 		for q in range(4):
134 | 			neta[p][q] = sum(u[p][q])/((sum(u[0][0]))**(((p+q)/2)+1))
135 | 	feature_vector.append(neta[2][0] + neta[0][2]) #feature 27
136 | 	feature_vector.append((neta[2][0] - neta[0][2])**2 + 4*((neta[1][1])**2)) #feature 28
137 | 	feature_vector.append((neta[3][0] - 3*neta[1][2])**2 + (3*neta[2][1] - neta[0][3])**2) #feature 29
138 | 	feature_vector.append((neta[3][0] + neta[1][2])**2 + (neta[2][1] + neta[0][3])**2) #feature 30
139 | 	feature_vector.append(((neta[3][0]-3*neta[1][2])*(neta[3][0] + neta[1][2])*((neta[3][0] + neta[1][2])**2 - 3*((neta[2][1] + neta[0][3])**2))) + ((3*neta[2][1]-neta[0][3])*(neta[2][1] + neta[0][3])*(3*(neta[3][0] + neta[1][2])**2 - ((neta[2][1] + neta[0][3])**2)))) #feature 31
140 | 	feature_vector.append((neta[2][0] - neta[0][2])*(((neta[3][0] + neta[1][2])**2) - ((neta[2][1] + neta[0][3])**2)) + (4*neta[1][1]*(neta[3][0] + neta[1][2])*(neta[2][1] + neta[0][3]))) #feature 32
141 | 	feature_vector.append(((3*neta[2][1]-neta[0][3])*(neta[3][0] + neta[1][2])*((neta[3][0] + neta[1][2])**2 - 3*((neta[2][1] + neta[0][3])**2))) - ((neta[3][0] + 3*neta[1][2])*(neta[2][1] + neta[0][3])*(3*(neta[3][0] + neta[1][2])**2 - ((neta[2][1] + neta[0][3])**2)))) #feature 33
142 | 	
143 | 	return feature_vector
144 | 	 
145 | def process_lung(img):
146 | 	r=30
147 | 	img=img[r:512-r,r:512-r]
148 | 	img_closed,img_filted,img_thresh,img_opened=preprocess(img)
149 | 	img_roied_left_lung,img_roied_right_lung=seg_cont(img_closed)
150 | 	feature_vector_left = feature_extract(img_roied_left_lung)
151 | 	feature_vector_right = 1#feature_extract(img_roied_right_lung)
152 | 	return feature_vector_left,feature_vector_right,img_roied_left_lung,img_roied_right_lung#img,img_closed,img_filted,img_thresh,img_opened,img_roied_left_lung,img_roied_right_lung,feature_vector
153 | 	
154 | def process_lung_test(img):
155 | 	r=30
156 | 	img=img[r:512-r,r:512-r]
157 | 	img_closed,img_filted,img_thresh,img_opened=preprocess(img)
158 | 	img_roied_left_lung,img_roied_right_lung=seg_cont(img_closed)
159 | 	feature_vector_left = feature_extract(img_roied_left_lung)
160 | 	feature_vector_right = 1#feature_extract(img_roied_right_lung)
161 | 
162 | 	cv2.imshow('img_filted',img_filted)
163 | 	cv2.waitKey()
164 | 	cv2.imshow('img_thresh',img_thresh)
165 | 	cv2.waitKey()
166 | 	cv2.imshow('img_opened',img_opened)
167 | 	cv2.waitKey()
168 | 	cv2.imshow('img_closed',img_closed)
169 | 	cv2.waitKey()
170 | 	cv2.imshow('img_roied_left_lung',img_roied_left_lung)
171 | 	cv2.waitKey()
172 | 	cv2.imshow('img_roied_right_lung',img_roied_right_lung)
173 | 	cv2.waitKey()
174 | 	cv2.destroyAllWindows()
175 | 	
176 | 	return feature_vector_left,feature_vector_right,img_roied_left_lung,img_roied_right_lung
177 | 	#img,img_closed,img_filted,img_thresh,img_opened,img_roied_left_lung,img_roied_right_lung,feature_vector
178 | ace care about it\
179 | 
180 | for i in [k]: #replace k by the list of images u need to train with... Image name format could differ.Take care about it.
181 | 	if not os.path.isfile('Left/canc ('+str(i)+').jpg'):
182 | 		continue
183 | 	print(i)
184 | 	img=cv2.imread('Left/canc ('+str(i)+').jpg',0)	
185 | 	####
186 | 	r=30
187 | 	img=img[r:512-r,r:512-r]
188 | 
189 | 	img_closed,img_filted,img_thresh,img_opened=preprocess(img)
190 | 	img_roied_left_lung,img_roied_right_lung=seg_cont(img_closed)
191 | 	####
192 | 	cv2.imshow('img_filted',img_filted)
193 | 	cv2.waitKey()
194 | 	cv2.imshow('img_thresh',img_thresh)
195 | 	cv2.waitKey()
196 | 	cv2.imshow('img_opened',img_opened)
197 | 	cv2.waitKey()
198 | 	cv2.imshow('img_closed',img_closed)
199 | 	cv2.waitKey()
200 | 	####
201 | 	cv2.imshow('img_roied_leftlung1',img_roied_left_lung)
202 | 	cv2.waitKey()
203 | 	cv2.imshow('img_roied_rightlung2',img_roied_right_lung)
204 | 	cv2.waitKey()
205 | 	cv2.destroyAllWindows()
206 | 	feature_vector = feature_extract(img_roied_left_lung)
207 | 	print(len(feature_vector))
208 | 	####
209 | 	features=process_lung(img)
210 | 	print(len(features))
211 | 
212 | 
213 | 


--------------------------------------------------------------------------------
/NeuralNetwork.py:
--------------------------------------------------------------------------------
1 | from sklearn.neural_network import MLPClassifier
2 | from datafile import dataset,dataset_output
3 | import pickle
4 | 
5 | clf = MLPClassifier(solver='lbfgs', alpha=1e-5,hidden_layer_sizes=(40, 10), random_state=1,learning_rate_init=0.001,max_iter=10000)
6 | clf.fit(dataset, dataset_output)
7 | 
8 | filename = 'finalized_model.sav'
9 | pickle.dump(clf, open(filename, 'wb'))


--------------------------------------------------------------------------------
/PredictCancer.py:
--------------------------------------------------------------------------------
 1 | import cv2
 2 | import lungc
 3 | import pickle
 4 | 
 5 | num=input("Enter test image number: ")
 6 | img=cv2.imread("test/test ("+num+").jpg",0)
 7 | left_feature,right_feature,img_left,img_right=lungc.process_lung_test(img)
 8 | 
 9 | 
10 | filename = 'finalized_model.sav'
11 | loaded_model = pickle.load(open(filename, 'rb'))
12 | result = loaded_model.predict([left_feature])
13 | print(result[0])
14 | if(result[0]):
15 | 	print("Tumour present")
16 | else:
17 | 	print("Normal Lung")
18 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # Lung-Cancer-Detection-Using-Python
 2 | ## Dataset
 3 | The Cancer Imaging Archive (TCIA)
 4 | http://www.cancerimagingarchive.net/
 5 | 
 6 | ## Code files 
 7 | Code have been written in Modular fashion
 8 | 
 9 | PredictCancer.py : Final program for testing a image
10 | 
11 | NeuralNetwork.py : MLP using SKlearn to learn the features and saving the Weights using pickle
12 | 
13 | LungCancerTrain.py : All Image processing techniques and code for training the model are written here
14 | 
15 | dataset_create.py : For making the folders of both positive and negative cases and naming the images in required format
16 | 
17 | Test Case images of both categories and added in the repository along with its terminal output for reference
18 | 
19 | ## Dependencies
20 | Python3 , OpenCV - cv2 , pickle , datafile libraries 
21 | 
22 | ## Output
23 | 
24 | Positive case
25 | 
26 | ![cancer image](https://user-images.githubusercontent.com/33830482/42348966-c8301910-80c8-11e8-9427-34fba3c0b84c.png)
27 | ![cancer terminal](https://user-images.githubusercontent.com/33830482/42348967-c92cb17a-80c8-11e8-82df-c31cba6ac42a.png)
28 | 
29 | Negative case
30 | 
31 | ![no cancer image](https://user-images.githubusercontent.com/33830482/42348968-c95ed394-80c8-11e8-8c2e-5f25a61f3ccd.png)
32 | ![no cancer terminal](https://user-images.githubusercontent.com/33830482/42348970-c9da984e-80c8-11e8-87e0-4afe7dde8bfb.png)
33 | 
34 | This work was done in partnership with my friend Tarun Bhargav Sriram as a project for Digital Image Processing elective.
35 | Any queries about the project, contact at vinaybn8997@gmail.com
36 | 


--------------------------------------------------------------------------------
/dataset_create.py:
--------------------------------------------------------------------------------
 1 | import lungc
 2 | import os.path
 3 | import cv2
 4 | import random
 5 | import numpy as np
 6 | 
 7 | nocanc=[]
 8 | nocanc_output =[]
 9 | for i in range(1,139):#without folder #run for full data
10 | 	if not os.path.isfile('without/nocanc ('+str(i)+').jpg'):
11 | 		continue
12 | 	print(i)
13 | 	img=cv2.imread('without/nocanc ('+str(i)+').jpg',0)
14 | 	left_feature,right_feature,img_left,img_right=lungc.process_lung(img)
15 | 	
16 | 	nocanc.append(left_feature)
17 | 	nocanc_output.append(0)
18 | 
19 | cancl=[]
20 | cancl_output =[]
21 | for i in range(1,121):#left folder #run for full data
22 | 	if not os.path.isfile('Left/canc ('+str(i)+').jpg'):
23 | 		continue
24 | 	print(i)
25 | 	img=cv2.imread('Left/canc ('+str(i)+').jpg',0)	
26 | 	left_feature,right_feature,img_left,img_right=lungc.process_lung(img)
27 | 	
28 | 	cancl.append(left_feature)
29 | 	cancl_output.append(1)
30 | 
31 | random.seed(2)
32 | dataset=cancl+nocanc
33 | dataset_output = cancl_output + nocanc_output 
34 | random.shuffle(dataset)
35 | random.shuffle(dataset_output)
36 | 
37 | with open('datafile.py', 'w') as f:
38 | 	f.write('dataset = %s' % dataset)
39 | 	f.write('\ndataset = %s' % dataset_output)
40 | 


--------------------------------------------------------------------------------