# -*- coding: utf-8 -*-
"""
Created on Wed Jan 20 14:09:32 2016
@author: Babak Ehteshami Bejnordi
Evaluation code for the Camelyon16 challenge on cancer metastases detecion
"""
import openslide
import numpy as np
import matplotlib.pyplot as plt
from scipy import ndimage as nd
from skimage import measure
import os
import sys
def computeEvaluationMask(maskDIR, resolution, level):
"""Computes the evaluation mask.
Args:
maskDIR: the directory of the ground truth mask
resolution: Pixel resolution of the image at level 0
level: The level at which the evaluation mask is made
Returns:
evaluation_mask
"""
slide = openslide.open_slide(maskDIR)
dims = slide.level_dimensions[level]
pixelarray = np.zeros(dims[0]*dims[1], dtype='uint')
pixelarray = np.array(slide.read_region((0,0), level, dims))
distance = nd.distance_transform_edt(255 - pixelarray[:,:,0])
Threshold = 75/(resolution * pow(2, level) * 2) # 75µm is the equivalent size of 5 tumor cells
binary = distance < Threshold
filled_image = nd.morphology.binary_fill_holes(binary)
evaluation_mask = measure.label(filled_image, connectivity = 2)
return evaluation_mask
def computeITCList(evaluation_mask, resolution, level):
"""Compute the list of labels containing Isolated Tumor Cells (ITC)
Description:
A region is considered ITC if its longest diameter is below 200µm.
As we expanded the annotations by 75µm, the major axis of the object
should be less than 275µm to be considered as ITC (Each pixel is
0.243µm*0.243µm in level 0). Therefore the major axis of the object
in level 5 should be less than 275/(2^5*0.243) = 35.36 pixels.
Args:
evaluation_mask: The evaluation mask
resolution: Pixel resolution of the image at level 0
level: The level at which the evaluation mask was made
Returns:
Isolated_Tumor_Cells: list of labels containing Isolated Tumor Cells
"""
max_label = np.amax(evaluation_mask)
properties = measure.regionprops(evaluation_mask)
Isolated_Tumor_Cells = []
threshold = 275/(resolution * pow(2, level))
for i in range(0, max_label):
if properties[i].major_axis_length < threshold:
Isolated_Tumor_Cells.append(i+1)
return Isolated_Tumor_Cells
def readCSVContent(csvDIR):
"""Reads the data inside CSV file
Args:
csvDIR: The directory including all the .csv files containing the results.
Note that the CSV files should have the same name as the original image
Returns:
Probs: list of the Probabilities of the detected lesions
Xcorr: list of X-coordinates of the lesions
Ycorr: list of Y-coordinates of the lesions
"""
Xcorr, Ycorr, Probs = ([] for i in range(3))
csv_lines = open(csvDIR,"r").readlines()
for i in range(len(csv_lines)):
line = csv_lines[i]
elems = line.rstrip().split(',')
Probs.append(float(elems[0]))
Xcorr.append(int(elems[1]))
Ycorr.append(int(elems[2]))
return Probs, Xcorr, Ycorr
def compute_FP_TP_Probs(Ycorr, Xcorr, Probs, is_tumor, evaluation_mask, Isolated_Tumor_Cells, level):
"""Generates true positive and false positive stats for the analyzed image
Args:
Probs: list of the Probabilities of the detected lesions
Xcorr: list of X-coordinates of the lesions
Ycorr: list of Y-coordinates of the lesions
is_tumor: A boolean variable which is one when the case cotains tumor
evaluation_mask: The evaluation mask
Isolated_Tumor_Cells: list of labels containing Isolated Tumor Cells
level: The level at which the evaluation mask was made
Returns:
FP_probs: A list containing the probabilities of the false positive detections
TP_probs: A list containing the probabilities of the True positive detections
NumberOfTumors: Number of Tumors in the image (excluding Isolate Tumor Cells)
detection_summary: A python dictionary object with keys that are the labels
of the lesions that should be detected (non-ITC tumors) and values
that contain detection details [confidence score, X-coordinate, Y-coordinate].
Lesions that are missed by the algorithm have an empty value.
FP_summary: A python dictionary object with keys that represent the
false positive finding number and values that contain detection
details [confidence score, X-coordinate, Y-coordinate].
"""
max_label = np.amax(evaluation_mask)
FP_probs = []
TP_probs = np.zeros((max_label,), dtype=np.float32)
detection_summary = {}
FP_summary = {}
for i in range(1,max_label+1):
if i not in Isolated_Tumor_Cells:
label = 'Label ' + str(i)
detection_summary[label] = []
FP_counter = 0
if (is_tumor):
for i in range(0,len(Xcorr)):
HittedLabel = evaluation_mask[int(Ycorr[i]/pow(2, level)), int(Xcorr[i]/pow(2, level))]
if HittedLabel == 0:
FP_probs.append(Probs[i])
key = 'FP ' + str(FP_counter)
FP_summary[key] = [Probs[i], Xcorr[i], Ycorr[i]]
FP_counter+=1
elif HittedLabel not in Isolated_Tumor_Cells:
if (Probs[i]>TP_probs[HittedLabel-1]):
label = 'Label ' + str(HittedLabel)
detection_summary[label] = [Probs[i], Xcorr[i], Ycorr[i]]
TP_probs[HittedLabel-1] = Probs[i]
else:
for i in range(0,len(Xcorr)):
FP_probs.append(Probs[i])
key = 'FP ' + str(FP_counter)
FP_summary[key] = [Probs[i], Xcorr[i], Ycorr[i]]
FP_counter+=1
num_of_tumors = max_label - len(Isolated_Tumor_Cells);
return FP_probs, TP_probs, num_of_tumors, detection_summary, FP_summary
def computeFROC(FROC_data):
"""Generates the data required for plotting the FROC curve
Args:
FROC_data: Contains the list of TPs, FPs, number of tumors in each image
Returns:
total_FPs: A list containing the average number of false positives
per image for different thresholds
total_sensitivity: A list containig overall sensitivity of the system
for different thresholds
"""
unlisted_FPs = [item for sublist in FROC_data[1] for item in sublist]
unlisted_TPs = [item for sublist in FROC_data[2] for item in sublist]
total_FPs, total_TPs = [], []
all_probs = sorted(set(unlisted_FPs + unlisted_TPs))
for Thresh in all_probs[1:]:
total_FPs.append((np.asarray(unlisted_FPs) >= Thresh).sum())
total_TPs.append((np.asarray(unlisted_TPs) >= Thresh).sum())
total_FPs.append(0)
total_TPs.append(0)
total_FPs = np.asarray(total_FPs)/float(len(FROC_data[0]))
total_sensitivity = np.asarray(total_TPs)/float(sum(FROC_data[3]))
return total_FPs, total_sensitivity
def plotFROC(total_FPs, total_sensitivity):
"""Plots the FROC curve
Args:
total_FPs: A list containing the average number of false positives
per image for different thresholds
total_sensitivity: A list containig overall sensitivity of the system
for different thresholds
Returns:
-
"""
fig = plt.figure()
plt.xlabel('Average Number of False Positives', fontsize=12)
plt.ylabel('Metastasis detection sensitivity', fontsize=12)
fig.suptitle('Free response receiver operating characteristic curve', fontsize=12)
plt.plot(total_FPs, total_sensitivity, '-', color='#000000')
plt.show()
if __name__ == "__main__":
mask_folder = sys.argv[1]
result_folder = sys.argv[2]
result_file_list = []
result_file_list += [each for each in os.listdir(result_folder) if each.endswith('.csv')]
EVALUATION_MASK_LEVEL = 5 # Image level at which the evaluation is done
L0_RESOLUTION = 0.243 # pixel resolution at level 0
FROC_data = np.zeros((4, len(result_file_list)), dtype=np.object)
FP_summary = np.zeros((2, len(result_file_list)), dtype=np.object)
detection_summary = np.zeros((2, len(result_file_list)), dtype=np.object)
ground_truth_test = []
ground_truth_test += [each[0:8] for each in os.listdir(mask_folder) if each.endswith('.tif')]
ground_truth_test = set(ground_truth_test)
caseNum = 0
for case in result_file_list:
print('Evaluating Performance on image:', case[0:-4])
sys.stdout.flush()
csvDIR = os.path.join(result_folder, case)
Probs, Xcorr, Ycorr = readCSVContent(csvDIR)
is_tumor = case[0:-4] in ground_truth_test
if (is_tumor):
maskDIR = os.path.join(mask_folder, case[0:-4]) + '.tif'
evaluation_mask = computeEvaluationMask(maskDIR, L0_RESOLUTION, EVALUATION_MASK_LEVEL)
ITC_labels = computeITCList(evaluation_mask, L0_RESOLUTION, EVALUATION_MASK_LEVEL)
else:
evaluation_mask = 0
ITC_labels = []
FROC_data[0][caseNum] = case
FP_summary[0][caseNum] = case
detection_summary[0][caseNum] = case
FROC_data[1][caseNum], FROC_data[2][caseNum], FROC_data[3][caseNum], detection_summary[1][caseNum], FP_summary[1][caseNum] = compute_FP_TP_Probs(Ycorr, Xcorr, Probs, is_tumor, evaluation_mask, ITC_labels, EVALUATION_MASK_LEVEL)
caseNum += 1
# Compute FROC curve
total_FPs, total_sensitivity = computeFROC(FROC_data)
# plot FROC curve
plotFROC(total_FPs, total_sensitivity)
eval_threshold = [.25, .5, 1, 2, 4, 8]
eval_TPs = np.interp(eval_threshold, total_FPs[::-1], total_sensitivity[::-1])
for i in range(len(eval_threshold)):
print('Avg FP = ', str(eval_threshold[i]))
print('Sensitivity = ', str(eval_TPs[i]))
print('Avg Sensivity = ', np.mean(eval_TPs))
