frida: an open source framework for image dataset analysis toby cornish, md, phd department of...

28
FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore, MD

Upload: linda-reed

Post on 25-Dec-2015

220 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

FrIDA:An open source framework for image

dataset analysis

Toby Cornish, MD, PhD

Department of Pathology

Johns Hopkins Medical Institutions

Baltimore, MD

Page 2: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

FrIDAFramework for Image Dataset Analysis

• Framework provides tools that allow users to create their own analysis methods

• Image Dataset iterates over a set of images, applying the same methods to each image

• Analysis returns per image statistics about area and intensity in a given region of interest

Page 3: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

FrIDA specs

• Written in Java

• Uses libraries from the popular Image J program (Wayne Rasband, NIH)

• Programmed by James Morgan

• Design by Toby Cornish, James Morgan and Angelo DeMarzo

Page 4: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

FrIDA specs, cont.

• Licensed under MPL and currently available at sourceforge:– http://sourceforge.net/projects/fridajhu/

• Requirements:– Java 6 runtime environment (JRE)

• Optional:– MySQL database

Page 5: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

FrIDA• Operates on 24 bit color image files, including TIFF,

JPEG, etc.

• Currently supports analysis of:– Area (total, ROI)– Intensity (mean, median, min, max)

• Saves all analysis settings to either an XML file or a MySQL database

• Results are saved to an XML file or plain text file

• Central concept: Masking…

Page 6: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking

• Mask:– a binary image that defines certain pixels

as included (foreground) or excluded (background)

Color Image “Tissue” Mask

Foreground = white

Background = black

Page 7: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking, cont.

• Grayscale segmentation is an example of a process that produces a mask

Color Image “Tissue” Mask

Grayscale segmentation

Page 8: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking, cont.

• FrIDA currently supports these masking methods:

– Color mask: uses HSB color space segmentation to select pixels within a color range

– Lasso mask: freehand selection of pixels using bounding polygons

– Metamask: combination of two masks using boolean logic

Page 9: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Main FrIDA window

Page 10: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking

• Color mask

• Lasso mask

• Metamask

Page 11: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Color space segmentation

• Segmentation assigning the pixels in an image to a particular category, i.e. classifying them

• Color segmentation grouping of pixels of similar color– The first step in any analysis of staining, i.e. which

pixels are brown and which are blue?

Page 12: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

HSB/HSV color space• More intuitive for human interaction• 1 pixel = 3 values

– Hue, Saturation, Brightness (Value)– H x S circle (chromaticity)

• Hue defines a color; saturation the amount of color present

– B (V) z-axis• Defines the “brightness”

S

H

0o

S

B

H

Page 13: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Color space segmentation

S

H = 0 to 35 degrees

S = 0 to 100%

I = 0 to 100% 0

o

35o

Page 14: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

HSB segmentation in FrIDA

• “Color Mask”

• Select colors by either:

– An eyedropper-style tool, and/or

– Three pairs of sliders that define bandpass ranges for Hue, Saturation and Brightness

Page 15: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Hue: Min > Max Bandstop

Page 16: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Brown

Color masking, cont.

HSB Segmentation

Page 17: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking

• Color mask

• Lasso mask

• Metamask

Page 18: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

“Lasso” masking

Freehand polygon tool for drawing regions of interest to be masked (red line)

Page 19: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

“Lasso” masking, cont.

Any number of lasso can be defined; multiple selections possible by holding down “shift”

Page 20: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

“Lasso” maskingRed lasso defines a mask

Exclude

Include

Page 21: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Masking

• Color mask

• Lasso mask

• Metamask

Page 22: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Meta masking• Metamasks combine two or more masks using

boolean logic:

– AND: the intersection of two masks• “all pixels in both mask1 AND mask2”

– OR: the union of two masks• “all pixels either in mask1 OR mask2”

– NOT: uses the inverse of a mask • “all pixels NOT in mask1”

Page 23: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Meta masking, cont.• Metamasks combine two or more masks using

boolean logic:

Page 24: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Meta masking, cont.

• Metamasks bring all the elements together for analysis…

Page 25: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

ExcludeInclude

Include AND (NOT Exclude)

NOT Exclude

NOT

Brown

AND

AND

Include AND (NOT Exclude) AND Brown

Page 26: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Include AND (NOT Exclude) AND Brown

Original

The mask is applied to the original image, selecting the pixels of interest and resultant image is analyzed

ANALYSIS

MASK

Masked Original

Page 27: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Future directions

• Preprocessing of images

• Additional masks:– e.g., binary operations

• Erosion, dilation, watershed segmentation, size exclusion, shape descriptor exclusion

• Additional results:– Particle counting

• Integration with TMAJ:– Display of masks while browsing TMAs

Page 28: FrIDA: An open source framework for image dataset analysis Toby Cornish, MD, PhD Department of Pathology Johns Hopkins Medical Institutions Baltimore,

Acknowledgements• James Morgan, B.S.• Bora Gurel, M.D.• Angelo DeMarzo, M.D., Ph.D.

Source for FrIDA can be obtained at sourceforge.net

Anyone interested in a live demo may request one during the meeting