1. # import libraries here
2. import math
3.  
4. import cv2
5. import numpy as np
6.  
7.  
8. def count_blood_cells(image_path):
9.     """
10.    Procedura prima putanju do fotografije i vraca broj crvenih krvnih zrnaca, belih krvnih zrnaca i
11.    informaciju da li pacijent ima leukemiju ili ne, na osnovu odnosa broja krvnih zrnaca
12.  
13.    Ova procedura se poziva automatski iz main procedure i taj deo kod nije potrebno menjati niti implementirati.
14.  
15.    :param image_path: <String> Putanja do ulazne fotografije.
16.    :return: <int>  Broj prebrojanih crvenih krvnih zrnaca,
17.             <int> broj prebrojanih belih krvnih zrnaca,
18.             <bool> da li pacijent ima leukemniju (True ili False)
19.    """
20.     red_blood_cell_count = 0
21.     white_blood_cell_count = 0
22.     has_leukemia = None
23.  
24.     # Ucitavanje slike
25.     img_bgr = cv2.imread(image_path)
26.  
27.     # TODO - Prebrojati crvena i bela krvna zrnca i vratiti njihov broj kao povratnu vrednost ove procedure
28.     white_blood_cell_count, wbc_contours, wbc_radius = count_white_cells(img_bgr)
29.     red_blood_cell_count = count_red_cells(img_bgr, wbc_contours, wbc_radius)
30.  
31.     # TODO - Odrediti da li na osnovu broja krvnih zrnaca pacijent ima leukemiju i vratiti True/False kao povratnu vrednost ove procedure
32.     has_leukemia = (white_blood_cell_count or 0) / (red_blood_cell_count or 1) > 0.085
33.  
34.     return red_blood_cell_count, white_blood_cell_count, has_leukemia
35.  
36.  
37. def count_white_cells(img_bgr):
38.     # Binary image from HSV
39.     img_hsv = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2HSV)
40.  
41.     min_blue = (50, 50, 20)
42.     max_blue = (145, 255, 255)
43.  
44.     img_hsw_filtered = cv2.inRange(img_hsv, min_blue, max_blue)
45.  
46.     # Binary image from threshold
47.     img_gray = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2GRAY)
48.  
49.     _, img_thresh_filtered = cv2.threshold(img_gray, 150, 255, cv2.THRESH_BINARY)
50.     img_thresh_filtered = cv2.bitwise_not(img_thresh_filtered)
51.  
52.     # Concatenate both images
53.     img = cv2.bitwise_or(img_hsw_filtered, img_thresh_filtered)
54.  
55.     # Fill holes
56.     kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
57.     img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
58.  
59.     _, contours, hierarchy = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
60.  
61.     contours_by_size = sorted(contours, key=cv2.contourArea, reverse=True)
62.  
63.     wbc_contours = []
64.     contours_to_remove = []
65.     min_area = 0
66.     for c in contours_by_size:
67.         area = cv2.contourArea(c)
68.         if area > min_area:
69.             wbc_contours.append(c)
70.             min_area = area / 3
71.         else:
72.             contours_to_remove.append(c)
73.  
74.     # Remove small contours
75.     for c in contours_to_remove:
76.         cv2.drawContours(img, [c], -1, 0, -1)
77.  
78.     if not wbc_contours:
79.         return 0, None, 0
80.  
81.     # Assume that last item in wbc_contours is a single cell
82.     wbc_area = cv2.contourArea(wbc_contours[-1])
83.     radius = np.round(2 * math.sqrt(wbc_area / math.pi))
84.  
85.     min_radius = np.int(radius * 0.3)
86.     max_radius = np.int(radius * 1.3)
87.  
88.     img_edges = cv2.Canny(img, 75, 200)
89.     circles = cv2.HoughCircles(img_edges, cv2.HOUGH_GRADIENT, 1.3, radius, param1=200, param2=30, minRadius=min_radius,
90.                                maxRadius=max_radius)
91.  
92.     wbc_count = 0
93.  
94.     if circles is not None:
95.         circles = np.round(circles[0, :]).astype("int")
96.         circles = sorted(circles, key=lambda x: x[2], reverse=True)
97.         biggest_radius = circles[0][2]
98.  
99.         for (x, y, r) in circles:
100.             for c in wbc_contours:
101.                 # If center is in contour or near contour
102.                 closes_contour_point = cv2.pointPolygonTest(c, (x, y), True)
103.                 if closes_contour_point >= 0 or -closes_contour_point < min_radius:
104.                     # cv2.circle(img_bgr, (x, y), r, (0, 255, 0), 2)
105.                     # cv2.rectangle(img_bgr, (x - 2, y - 2), (x + 2, y + 2), (0, 128, 255), -1)
106.                     wbc_count += np.int(biggest_radius / r)
107.                     break
108.  
109.     print(wbc_count)
110.  
111.     return wbc_count, wbc_contours, radius
112.  
113.  
114. def count_red_cells(img_bgr, wbc_contours, wbc_radius):
115.     img_gray = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2GRAY)
116.     img = cv2.medianBlur(img_gray, 5)
117.     img = cv2.GaussianBlur(img, (5, 5), 0)
118.     img = cv2.equalizeHist(img)
119.  
120.     img = cv2.adaptiveThreshold(img_gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 41, 5)
121.     img = cv2.bitwise_not(img)
122.  
123.     # Remove wbc countours
124.     for c in wbc_contours:
125.         cv2.drawContours(img, [c], -1, 0, -1)
126.  
127.     kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
128.  
129.     _, contours, _ = cv2.findContours(img, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
130.     for c in contours:
131.         cv2.drawContours(img, [c], 0, 255, -1)
132.  
133.     _, contours, _ = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
134.  
135.     rbc_radii = []
136.     for c in contours:
137.         if len(c) >= 5:
138.             ellipse = cv2.fitEllipse(c)
139.             widthE = ellipse[1][0]
140.             heightE = ellipse[1][1]
141.             if widthE > heightE:
142.                 coef = heightE / widthE
143.             else:
144.                 coef = widthE / heightE
145.  
146.             if coef > 0.8:
147.                 (x, y), r = cv2.minEnclosingCircle(c)
148.                 rbc_radii.append(r)
149.  
150.     min_radius = 0
151.     similar_radii = []
152.     for r in sorted(rbc_radii, reverse=True):
153.         if r > min_radius:
154.             min_radius = r * 0.8
155.             similar_radii.append(r)
156.  
157.     radius = np.average(similar_radii)
158.  
159.     min_radius = np.int(radius * 0.6)
160.  
161.     # Detect
162.     rbc_contours = []
163.     rbc_count = 0
164.     for c in contours:
165.         (x, y), r = cv2.minEnclosingCircle(c)
166.         if r > min_radius:
167.             cv2.circle(img_bgr, (int(x), int(y)), int(r), (0, 255, 0), 2)
168.             rbc_contours.append(c)
169.             rbc_count += int(r / min_radius)
170.  
171.     return len(rbc_contours)
172.