mistycheney

# Install pyenv
export PYENV_ROOT=/nfs/intern_data/thsu/.pyenv
export PATH="${PYENV_ROOT}/bin:${PATH}"
curl https://pyenv.run | bash

# Then add in ~/.bashrc:
eval "$(pyenv init --path)"
eval "$(pyenv init -)"
eval "$(pyenv virtualenv-init -)"

mistycheney

import pyemd # https://github.com/garydoranjr/pyemd. Install by running 'pip install pyemd'.
from scipy.spatial.distance import pdist, squareform

x_bins = range(10) # Assume histogram has 10 bins for each axis
y_bins = range(10)
z_bins = range(10)

xs, ys, zs = np.meshgrid(x_bins, y_bins, z_bins)
xs = xs.flatten()
ys = ys.flatten()

mistycheney

# Various ways to find peaks in Python
# https://github.com/MonsieurV/py-findpeaks

def count_anomalies_from_scores(scores, plot=False):
    from scipy.signal import find_peaks
    is_outlier = detect_outlier(scores)
    peaks, _ = find_peaks(is_outlier, distance=60) # 60 means peaks must be separated by 60 mins
    num_anomalies = np.count_nonzero(peaks)
    
    if plot:

mistycheney

// Batch Convert
//
// This macro convert all the files in a folder to TIFF, 8-bit TIFF, 
// JPEG, GIF, PNG, PGM, BMP, FITS, Text Image, ZIP or Raw
// format. Three dialog boxes are displayed. Select the source 
// folder in the first, the format in the second and the destination 
// folder in the third. Batch_Converter, a similar plugin is at 
//    http://rsb.info.nih.gov/ij/plugins/batch-converter.html
// https://imagej.nih.gov/ij/plugins/batch-converter.html

mistycheney

mistycheney

Incorporating Uncertainty

If the score maps given by the classifier are not extremely reliable, we need to balance between the reference structure locations constrained by the atlas and the evidence from score maps.

We do this in a Bayesian setting. Suppose the atlas gives a prior distribution for the structure's location, represented by $P(\theta)$. An observation distribution $P(s | \Omega)$ models classification uncertainty, where $\Omega$ is the set of voxels that belong to a particular structure, and $s$ is the computed score volume.

Let $\Omega(\theta)$ denote this structure's voxel locations transformed by $\theta$, and let $s$ denote the score volume, then the likelihood of the score volume is $P(s | \Omega(\theta))$. The optimal transform $\theta^*$ should maximize the posterior:

$$ \theta^* = \arg\max P(s | \Omega(\theta)) P(\theta).$$

mistycheney

# http://stackoverflow.com/a/31402351

def bbox_2d(img):
    rows = np.any(img, axis=1)
    cols = np.any(img, axis=0)
    rmin, rmax = np.where(rows)[0][[0, -1]]
    cmin, cmax = np.where(cols)[0][[0, -1]]

    return cmin, cmax, rmin, rmax

mistycheney

# https://gist.github.com/hellpanderrr/2c08af0f07eed4782234

from scipy.spatial import Delaunay, ConvexHull
import networkx as nx

def concave_hull(points,alpha_x=150,alpha_y=250):
    points = [(i[0],i[1]) if type(i) <> tuple else i for i in points]
    de = Delaunay(points)
    dec = []
    a = alpha_x

mistycheney

def less(center):
    def less_helper(a, b):
        if (a[0] - center[0] >= 0 and b[0] - center[0] < 0):
            return 1;
        if (a[0] - center[0] < 0 and b[0] - center[0] >= 0):
            return -1;
        if (a[0] - center[0] == 0 and b[0] - center[0] == 0):
            if (a[1] - center[1] >= 0 or b[1] - center[1] >= 0):
                return 2*int(a[1] > b[1]) - 1;
            return 2*int(b[1] > a[1]) - 1

mistycheney

stack = 'MD589'
data_dir = '/oasis/projects/nsf/csd395/yuncong/CSHL_data_patches/%s_byLandmark'%(stack)

all_image_names = []
for label in os.listdir(data_dir):
    if label == 'BackG':
        # limit the number of BackG examples
        for img_filename in os.listdir(data_dir + '/' + label)[:300]:
            all_image_names.append((label_dict[label], label + '/' + img_filename))
    else: