Skip to content

Commit

Permalink
NC Ratio: analysis now runs in about 1 sec.
Browse files Browse the repository at this point in the history 
Speedup mainly by re-using pre-allocated memory and by skipping
the slow gaussian blur to calculate a background image
  • Loading branch information
Nico Stuurman committed Dec 4, 2019
1 parent a75cc42 commit ba5fb0d
Showing 1 changed file with 54 additions and 24 deletions.
78 changes: 54 additions & 24 deletions MicroNuclei/src/org/micromanager/micronuclei/analysis/NucleoCytoplasmicRatio.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -89,13 +89,17 @@ public class NucleoCytoplasmicRatio extends AnalysisModule {
private TextWindow textWindow_;

// Intermediate BoofCV images. Keep references for efficiency
private GrayU16 nuclearImgBackground_;
private GrayS32 backgroundSubtractedNuclearImg;
private GrayU16 gaussNuc_;
private GrayS32 contourImg_;
private GrayU16 cytoBlurred_;
private GrayU8 cytoMask_;

// since our class is recreated every run, they need to be static
private static GrayU16 nuclearImgBackground_;
private static GrayS32 backgroundSubtractedNuclearImg;
private static GrayU16 gaussNuc_;
private static GrayS32 contourImg_;
private static GrayU16 cytoBlurred_;
private static GrayU8 cytoMask_;
private static GrayU8 thresholdedNuc_;
private static GrayU8 thresholdedNuc2_;


public NucleoCytoplasmicRatio() {
// note: the type of the value when creating the AnalysisProperty determines
// the allowed type, and can create problems when the user enters something
Expand Down Expand Up @@ -198,6 +202,9 @@ public ResultRois analyze(Studio mm, Image[] imgs, Roi userRoi, JSONObject parms
final double pixelSurface = nuclImg.getMetadata().getPixelSizeUm() *
nuclImg.getMetadata().getPixelSizeUm();


long start = System.currentTimeMillis();

GrayU16 bNuc = (GrayU16) BoofCVImageConverter.mmToBoofCV(nuclImg, false);
ImageGray igCyto = BoofCVImageConverter.mmToBoofCV(cytoImg, false);
nuclearImgBackground_ = (GrayU16) createSameShapeIfNeeded(bNuc,
Expand All @@ -208,32 +215,33 @@ public ResultRois analyze(Studio mm, Image[] imgs, Roi userRoi, JSONObject parms
contourImg_ = (GrayS32) createSameShapeIfNeeded(bNuc, contourImg_, GrayS32.class);

// Heavy blur to create a "background" image
/* Skip because it is slow
GBlurImageOps.gaussian(bNuc, nuclearImgBackground_, -1, 400, null);
// subtract the minimum from background to avoid values < 0 after subtraction
final int nuclearBackgroundMinimum = ImageStatistics.min(nuclearImgBackground_);
PixelMath.minus(nuclearImgBackground_, nuclearBackgroundMinimum, nuclearImgBackground_);
// Subtract background from nuclear image
PixelMath.subtract(bNuc, nuclearImgBackground_, backgroundSubtractedNuclearImg);
ConvertImage.convert(backgroundSubtractedNuclearImg, bNuc);
*/
GBlurImageOps.gaussian(bNuc, gaussNuc_, -1, 3, null);

int otsu2 = GThresholdImageOps.computeOtsu2(gaussNuc_, ImageStatistics.min(gaussNuc_),
ImageStatistics.max(gaussNuc_));
GrayU8 thresholdedNuc = gaussNuc_.createSameShape(GrayU8.class);
GrayU8 thresholdedNuc2 = gaussNuc_.createSameShape(GrayU8.class);

GThresholdImageOps.threshold(gaussNuc_, thresholdedNuc, otsu2, false);
thresholdedNuc_ = (GrayU8) createSameShapeIfNeeded(gaussNuc_, thresholdedNuc_, GrayU8.class);
thresholdedNuc2_ = (GrayU8) createSameShapeIfNeeded(gaussNuc_, thresholdedNuc2_, GrayU8.class);
GThresholdImageOps.threshold(gaussNuc_, thresholdedNuc_, otsu2, false);

// close using dilate/erode
BinaryImageOps.dilate4(thresholdedNuc, 2, thresholdedNuc2);
BinaryImageOps.erode4(thresholdedNuc2, 2, thresholdedNuc);
BinaryImageOps.dilate4(thresholdedNuc_, 2, thresholdedNuc2_);
BinaryImageOps.erode4(thresholdedNuc2_, 2, thresholdedNuc_);

// Use ImageJ Watershed code. Worth porting to BoofCV?
ImageProcessor d = BoofCVImageConverter.convert(thresholdedNuc, false);
ImageProcessor d = BoofCVImageConverter.convert(thresholdedNuc_, false);
EDM edm = new EDM();
edm.toWatershed(d);
// note: d shares pixels with thesholdedNuc, so we can erode thresholdedNuc directly!
BinaryImageOps.erode4(thresholdedNuc, 2, thresholdedNuc2);
BinaryImageOps.erode4(thresholdedNuc_, 2, thresholdedNuc2_);


/*
Expand All @@ -258,9 +266,15 @@ public ResultRois analyze(Studio mm, Image[] imgs, Roi userRoi, JSONObject parms
IJ.run(nuclIp, "Options...", "iterations=1 count=1 black pad edm=Overwrite do=Erode");
ImageGray igNuc = BoofCVImageConverter.convert(nuclIp.getProcessor(), false);
*/


long endImgAnalysis = System.currentTimeMillis();
long duration = endImgAnalysis - start;
System.out.println("Img analysis: " + duration + " msec");


List<Contour> contours =
BinaryImageOps.contour(thresholdedNuc2, ConnectRule.FOUR, contourImg_);
BinaryImageOps.contour(thresholdedNuc2_, ConnectRule.FOUR, contourImg_);
List<List<Point2D_I32>> tmpNuclearClusters =
BinaryImageOps.labelToClusters(contourImg_, contours.size(), null);
Map<Integer, List<Point2D_I32>> nuclearClusters = new HashMap<>(tmpNuclearClusters.size());
Expand All @@ -277,14 +291,14 @@ public ResultRois analyze(Studio mm, Image[] imgs, Roi userRoi, JSONObject parms
// this defines an "empty" ring between nucleus and cytoplasm
for (int i = 0; i < 3; i++) {
expandedNuclearCluster =
BinaryListOps.dilate4_2D_I32(expandedNuclearCluster, thresholdedNuc2.width, thresholdedNuc2.height);
BinaryListOps.dilate4_2D_I32(expandedNuclearCluster, thresholdedNuc2_.width, thresholdedNuc2_.height);
}
Set<Point2D_I32>cytoCluster =
BinaryListOps.dilate4_2D_I32(expandedNuclearCluster, thresholdedNuc2.width, thresholdedNuc2.height);
BinaryListOps.dilate4_2D_I32(expandedNuclearCluster, thresholdedNuc2_.width, thresholdedNuc2_.height);
// this defines the "thickness" of the cytoplasmic ring
for (int i = 0; i < 4; i++) {
cytoCluster =
BinaryListOps.dilate4_2D_I32(cytoCluster, thresholdedNuc2.width, thresholdedNuc2.height);
BinaryListOps.dilate4_2D_I32(cytoCluster, thresholdedNuc2_.width, thresholdedNuc2_.height);
}
cellClusters.put(index, cytoCluster);
cytoCluster = BinaryListOps.subtract(cytoCluster, expandedNuclearCluster);
Expand Down Expand Up @@ -379,19 +393,35 @@ public ResultRois analyze(Studio mm, Image[] imgs, Roi userRoi, JSONObject parms
final double nuclearSize = nucleus.size() * pixelSurface;
final double cytoSize = cyto.size() * pixelSurface;
if (nuclearSize > (double) minSizeN_.get() && nuclearSize < (double) maxSizeN_.get()) {
textWindow_.append(pos + "-" + counter++ + "\t" + (int) centers.get(i).x + "\t"
+ (int) centers.get(i).y + "\t" + nuclearSize + "\t" + nuclearAvg + "\t"
+ nAvg + "\t" + cytoSize + "\t" + cAvg + "\t" + nAvg / cAvg + "\n");
StringBuilder sb = new StringBuilder();
sb.append(pos).append( "-").append(counter).append("\t").
append((int) centers.get(i).x).append("\t").
append((int) centers.get(i).y).append("\t").
append(nuclearSize).append("\t").
append(nuclearAvg).append("\t").
append(nAvg).append("\t").
append(cytoSize).append("\t").
append(cAvg).append("\t").
append(nAvg / cAvg).append("\n");
textWindow_.append(sb.toString());
}
}



long endAnalysis = System.currentTimeMillis();
duration = endAnalysis - endImgAnalysis;

System.out.println("Analysis: " + duration + " msec");




/**
* Uncomment to display the nuclear and cytoplasmic masks
* Display nuclear and cytoplasmic masks if desired
*/
if ((Boolean) showMasks_.get()) {
GrayU8 dispImg = new GrayU8(thresholdedNuc2.getWidth(), thresholdedNuc2.getHeight());
GrayU8 dispImg = new GrayU8(thresholdedNuc2_.getWidth(), thresholdedNuc2_.getHeight());
for (int i = 0; i < nuclearClusters.size(); i++) {
List<Point2D_I32> cluster = nuclearClusters.get(i);
for (Point2D_I32 p : cluster) {
Expand Down

0 comments on commit ba5fb0d

Please sign in to comment.