milestone_03012018_FINAL.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluating and Optimizing Machine Learning Techniques for Automatic Nuclei Detection\n",
    "\n",
    "<b>1. Motivation</b>\n",
    "\n",
    "Many people die from cancer every year. This is terribly sad. :’( We would like to prevent this. One option is to identify the cancer early; treatment of nascent cancers tends to be more successful. Often, cells suffering from cancer display different biomarkers in their nuclei. These biomarkers can be seen in immunohistochemical (IHC) imaging. If there were a rapid and accurate way to screen images for cancerous cells, many cancer prognoses could be improved. We aim to develop an automated cell nuclei detection technique.\n",
    "\n",
    "<b>2. Challenges</b>\n",
    "\n",
    "Imaging conditions can vary tremendously - illumination, contrast, fluorescence and staining will all affect the appearance of the cell. We would also like our technique to be generalizable to many cell types because cancer is insidious and affects many cell types. Depending on the biological sample that was imaged, cells may also be aggregated; where a trained eye could distinguish individual nuclei, an algorithm might falter. \n",
    "\n",
    "<b>3. Pre-treatment and Initial Segmentation (“Detection”)</b>\n",
    "\n",
    "Colour is usually either normalized or thresholded to remove noise and background [1]. A variety of pre-processing techniques can be used to find objects, from conventional blob detection, to morphological and/or contour resolution, to watershed segmentation [2-4]. We will use sensible discretion to pick a pre-treatment technique.\n",
    "\n",
    "<b>4. Segmentation and Identification</b>\n",
    "\n",
    "There are two primary approaches to cell segmentation:\n",
    "\n",
    "<b><i>Traditional methods segment nuclei from single or overlapping cells</i></b>\n",
    "\n",
    "Several traditional methods, often involving a-priori knowledge of cell shape and size [5]. First, cell clusters are segmented from the background by concavity [6-7]. Next, cell clusters can be separated into individual cells based on the concavity of the intensity distribution [7]. After individual cells have been identified, cell boundaries are often approximated using elliptical curve-fitting techniques [5,7]. Further segmentation can be applied to separate the cell nucleus from the cytoplasm. A gradient vector flow active contour model (GVF-ACM) has been shown to find boundaries between the nucleus and cytoplasm [8].\n",
    "\n",
    "In this work, we can begin performing segmentation using the scikit-image package for python. This package includes methods for ellipse and boundary fitting, as well as edge detection and active contour modeling.\n",
    "\n",
    "<b><i>Machine learning methods identify nuclei via classification algorithms</i></b>\n",
    "\n",
    "Machine learning and pattern recognition have been successfully used to identify and segment cells in IHC images [9]. Whether using techniques such as cluster analysis [10], random forests [9], or deep neural networks [11-13], the workflow is similar. First, initial segmentation is performed (often in the pre-treatment step) to find cells and agglomerates. Then, training data is fed through a classifier to extract the most important features. Repeating this process while keeping only the most important feature vectors establishes a model, which is finally used to classify new test data.\n",
    "\n",
    "We will start with cluster analysis and random forest classifiers (from the scikit-learn package), but likely will also employ traditional GVF or ellipse-fitting algorithms (which we will implement in python) to refine initial segmentation. If necessary, we also intend to look into convolutional neural networks implemented with the TensorFlow package.\n",
    "\n",
    "### References\n",
    "\n",
    "[1] M. Veta, P.J. Van Diest, R. Kornegoor, A. Huisman, M.A. Viergever, and J.P.W. Pluim. (<b>2013</b>) Automatic Nuclei Segmentation in H&E Stained Breast Cancer Histopathology Images. <i>PLoS One</i>, 8, 7.\n",
    "\n",
    "[2] X. Yang, H. Li, and X. Zhou. (<b>2006</b>) Nuclei Segmentation using Marker-Controlled Watershed, Tracking using Mean-Shift, and Kalman Filter in Time-Lapse Microscopy. <i>IEEE Trans. Circuits Syst. I, Reg. Papers.</i>, 53, 11.\n",
    "\n",
    "[3] J. Cheng and J.C. Rajapakse. (<b>2009</b>) Segmentation of Clustered Nuclei with Shape Markers and Marking Function. <i>IEEE Trans. Biomed. Eng.</i>, 56, 3.\n",
    "\n",
    "[4] S. Ali and A. Madabhushi, “An Integrated Region-, Boundary-, Shape-Based Active Contour for Multiple Object Overlap Resolution in Histological Imagery. <i>IEEE Trans. Med. Imag.</i>, 31, 7.\n",
    "\n",
    "[5] S. Kothari, Q. Chaudry, M.D. Wang. (2009) Automated Cell Counting and Cluster Segmentation Using Concavity Detection and Ellipse Fitting Techniques. <i>Proc. IEEE Int. Symp. Biomed. Imaging</i>, 795.\n",
    "\n",
    "[6] Y. Toyoshima, T. Tokunaga, O. Hirose, M. Kanamori, T. Teramoto, M.S. Jang, S. Kuge, T. \n",
    "Ishihara, R. Yoshida, and Y. Iino. (<b>2006</b>) Accurate Automatic Detection of Densely Distributed Cell Nuclei in 3D Space. <i>PLoS. Comput. Biol.</i>, 12, 6.\n",
    "\n",
    "[7] H.S. Wu, J. Gil, and J.Barba. (<b>1998</b>) Optimal Segmentation of Cell Images. <i>IEE P-Vis. Image. Sign.</i>, 145, 1.\n",
    "\n",
    "[8] S.F. Yang-Mao, Y.K. Chan, and Y.P. Chu. (<b>2008</b>) Edge enhancement nucleus and cytoplast contour detector of cervical smear images. <i>IEEE Trans. Syst. Man, Cybern. B</i>, 38, 2.\n",
    "\n",
    "[9] O. Rujuta and A.J. Vyavahare. (<b>2017</b>) Review of Nuclei Detection, Segmentation in Microscopic Images. <i>J. Bioengineer. Biomed. Sci.</i>, 7, 2.\n",
    "\n",
    "[10] S. Wienert, D. Heim, K. Saeger, A. Stenzinger, M. Beil, P. Hufnagl, M. Dietel, C. Denkert, F. Klauschen. (<b>2012</b>) Detection and Segmentation of Cell Nuclei in Virtual Microscopy Images: A Minimum-Model Approach. <i>Sci. Rep.</i>, 2, 503.\n",
    "\n",
    "[11] S.K. Sadanandan, P. Ranefall, S. Le Guyader, and C. Wahlby. (<b>2017</b>) Automated Training of Deep Convolutional Neural Networks for Cell Segmentation. <i>Sci. Rep.</i>, 7, 1.\n",
    "\n",
    "[12] K. Sirinukunwattana, S.E.A. Raza, Y.W Tsang, I.A. Cree, D.R.J. Snead, and N.M. Rajpoo. (<b>2016</b>) Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images. <i>IEEE Trans. Med. Imag.</i>, 35, 99.\n",
    "\n",
    "[13] N. Kumar, R. Verma, S. Sharma, S. Bhargava, A. Vahadane, and A. Sethi. (<b>2017</b>) A Dataset and a Technique for Generalized Nuclear Segmentation for Computational Pathology. <i>IEEE Trans. Med. Imag.</i>, 36, 7."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## change the following to where you have stored and/or extracted the files\n",
    "# path = './project/'\n",
    "path = '/Users/arrakis/Dropbox/Tool - Classes/ChBE 8803/Project'\n",
    "\n",
    "n_samples = 3\n",
    "#n_samples = 560 # 560 samples to train on\n",
    "\n",
    "###################################################################"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## load all packages used below\n",
    "from skimage.color import rgb2gray\n",
    "from skimage.filters import threshold_otsu\n",
    "import imageio\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import zipfile,io\n",
    "import pylab as plt\n",
    "import sklearn\n",
    "import matplotlib.image as mpimg\n",
    "from scipy import ndimage\n",
    "from sklearn import linear_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQYAAAD8CAYAAACVSwr3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1c227c50b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "## STEP 1: Load an image (by index) and corresponding masks from ZIPPED stage1_train as np array\n",
    "\n",
    "def load_zipped_img(path, img_index): # load an image and all its masks\n",
    "    z = zipfile.ZipFile(path,'r') # access zip folder\n",
    "    zlist = z.namelist() # list of files in zip folder\n",
    "    img_name = zlist[img_index] # get selected image\n",
    "    img_name = img_name[0:-1] # eliminate \"/\"\n",
    "    \n",
    "    # get image and return as np array\n",
    "    img_raw = z.read('{}/images/{}.png'.format(img_name,img_name)) # get raw image\n",
    "    img = io.BytesIO(img_raw) # convert image\n",
    "    img = mpimg.imread(img) # numpy array\n",
    "    img = np.flip(img,0) # flip image\n",
    "    \n",
    "    # get all masks and return as np array\n",
    "    mask_list = []\n",
    "    for string in zlist:\n",
    "        if string.startswith(img_name+'/mask'):\n",
    "            mask_list.append(string)\n",
    "    mask_list = mask_list[1:-1] # list of masks\n",
    "    \n",
    "    masks = []\n",
    "    for m in mask_list:\n",
    "        mask_raw = z.read(m) # get raw mask\n",
    "        mask = io.BytesIO(mask_raw) # convert mask\n",
    "        mask = mpimg.imread(mask) # numpy array\n",
    "        mask = np.flip(mask,0) # flip mask\n",
    "        masks.append(mask)\n",
    "        \n",
    "    return img, masks\n",
    "\n",
    "# WORKING EXAMPLE OF load_zipped_img\n",
    "(img, masks) = load_zipped_img(path+'/stage1_train.zip',1)\n",
    "imgplot = plt.imshow(img, origin='lower')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQYAAAD8CAYAAACVSwr3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMytxMJhZyf8D8Tm4SniUoQ4AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1139fc748>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "## STEP 2: Grayscale image and segment objects by thresholding images (currently using Otsu's method)\n",
    "\n",
    "# convert to grayscale\n",
    "def grayscale(im):\n",
    "    return rgb2gray(im)\n",
    "\n",
    "# Otsu's Method, calculates optimal threshold for equal inter-/intra-class variance\n",
    "def otsu(image_gray):\n",
    "    threshold_val = threshold_otsu(image_gray) #Select threshold from Otsu's method\n",
    "    img_masked = np.where(image_gray > threshold_val, 1, 0)\n",
    "\n",
    "    if np.sum(img_masked==0) < np.sum(img_masked==1):\n",
    "        img_masked = np.where(img_masked, 0, 1)\n",
    "    return img_masked\n",
    "\n",
    "image_gray = grayscale(img)\n",
    "img_masked = otsu(image_gray)\n",
    "imgplot = plt.imshow(img_masked, origin='lower')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## STEP 3: Separate individual objects and encode in run-length format\n",
    "\n",
    "# separate objects in image into individual masks\n",
    "def separate_obj(img_masked):\n",
    "    labels, nlabels = ndimage.label(img_masked)\n",
    "\n",
    "    label_arrays = []\n",
    "    for label_num in range(1, nlabels+1):\n",
    "        label_mask = np.where(labels == label_num, 1, 0)\n",
    "        label_arrays.append(label_mask)\n",
    "    return labels, nlabels, label_mask\n",
    "\n",
    "# convert path to run-length encoding (RLE) output format\n",
    "def convert2runlength(x):\n",
    "    obj = np.where(x.T.flatten()==1)[0] #1 corresponds to object, 0 to background\n",
    "    run_lengths = []\n",
    "    prev = -2\n",
    "    for b in obj: # find continuous set of object pixels\n",
    "        if (b>prev+1): run_lengths.extend((b+1, 0))\n",
    "        run_lengths[-1] += 1\n",
    "        prev = b\n",
    "    return \" \".join([str(i) for i in run_lengths])\n",
    "\n",
    "def rle(img_masked, im_id):\n",
    "    (labels, nlabels, label_mask) = separate_obj(img_masked)\n",
    "    im_df = pd.DataFrame()\n",
    "    for label_num in range(1, nlabels+1):\n",
    "        label_mask = np.where(labels == label_num, 1, 0)\n",
    "        if label_mask.flatten().sum() > 10:\n",
    "            rle = convert2runlength(label_mask)\n",
    "            s = pd.Series({'ImageId': im_id, 'EncodedPixels': rle})\n",
    "            im_df = im_df.append(s, ignore_index=True)\n",
    "    return im_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LinearRegression(copy_X=True, fit_intercept=True, n_jobs=1, normalize=False)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# one-indexes a 2d array into 1d, top down then left right, output is np 1d array\n",
    "def one_index(arr2d):\n",
    "    h = arr2d.shape[0]\n",
    "    w = arr2d.shape[1]\n",
    "    \n",
    "    arr1d = []\n",
    "    for col in range(0, w):\n",
    "        for row in range(0, h):\n",
    "            arr1d.append(arr2d[row][col])\n",
    "    return np.array(arr1d)\n",
    "    \n",
    "# pads all vectors in array to have max_len, returns np array\n",
    "def pad_normalize(array, max_len):\n",
    "    for i in range(0, len(array)):\n",
    "        vec = array[i]\n",
    "        if len(vec) < max_len:\n",
    "            array[i] = np.concatenate(( np.array(vec).reshape(1,-1), np.zeros((1, (max_len-len(vec)))) ), axis=1)\n",
    "        else:\n",
    "            array[i] = np.array(vec).reshape(1,-1)\n",
    "    return np.array(array)\n",
    "\n",
    "## EXAMPLE TRAINING\n",
    "\n",
    "z = zipfile.ZipFile(path+'/stage1_train.zip','r') # access zip folder\n",
    "zlist = z.namelist() # list of files in zip directory\n",
    "samples = zlist[0:n_samples-1] # 0 < samples <= 560\n",
    "\n",
    "x_train = [] # predicted segmentation using Otsu's thresholding\n",
    "y_train = [] # \"correct\" segmentation from sum of masks\n",
    "max_len = 0\n",
    "for i in range(0, n_samples):\n",
    "    (img, masks) = load_zipped_img(path+'/stage1_train.zip', i) # loads image and associated masks\n",
    "    h = img.shape[0]\n",
    "    w = img.shape[1]\n",
    "    \n",
    "    x_vec = one_index(otsu(grayscale(img)))\n",
    "    y_vec = one_index(sum(masks))\n",
    "            \n",
    "    if len(x_vec) > max_len: max_len = len(x_vec)\n",
    "    x_train.append(x_vec)\n",
    "    y_train.append(y_vec)\n",
    "\n",
    "x_train = np.squeeze(pad_normalize(x_train, max_len), axis=1)\n",
    "x_shortfeature = x_train[:, 0:10000]\n",
    "y_train = np.squeeze(pad_normalize(y_train, max_len), axis=1)\n",
    "y_shortfeature = y_train[:, 0:10000]\n",
    "\n",
    "linreg = linear_model.LinearRegression() # create linear regression object\n",
    "linreg.fit(x_shortfeature, y_shortfeature) # train the model using the training sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7518\n"
     ]
    }
   ],
   "source": [
    "## EXAMPLE TESTING\n",
    "\n",
    "(img_test, mask_test) = load_zipped_img(path+'/stage1_train.zip', 5)\n",
    "X_test = one_index(otsu(grayscale(img)))\n",
    "X_test = X_test[:10000].reshape(1, -1)\n",
    "y_pred = linreg.predict(X_test) # predict using the testing set\n",
    "y_pred = sum(y_pred).reshape(1,-1)\n",
    "y_pred = np.round(y_pred)\n",
    "\n",
    "# Scoring\n",
    "y_actual = one_index(sum(mask_test))[:10000].reshape(1, -1)\n",
    "accuracy = sklearn.metrics.accuracy_score(y_pred.T,y_actual.T)\n",
    "\n",
    "print(accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}