mammography using matlab

Replies

• 

  Rupam Das

  I Have implemented it, using GLCM, Wavelet Transform, SVM, Level Snake Segmentation. I definitely mind sharing the code for free, But never mind for the Idea. You ask, I will guide........ Its not that tough You can Implement.
  
  By the way what database you are using?
• 

  oooivanooo

  Sorry for late reply, because busy on my personal stuff.
  
  
  Below are my processing flow and problems I have facing:
  1. Data collection: MIAS database
  
  2. Image preprocessing: Morphology
  Questions:
  a. Am I using the right preprocessing method? Why you are using GLCM? Is it more accurate compare to others?
  b. Do I need to crop the images manually before preprocessing since all the images are bias and not consistent in size?
  
  3. Image Analysis:
  i. Feature Extraction: STFT
  Questions:
  a. Am I using the correct method? Any other suitable method I can try? I'm not going to use wavelet transform because this method already used by other student.
  b. How to implement this step? As I mentioned in the step 2Qb, all the images are not consistent, so how to get the accurate result?
  
  ii. Feature Selection: Standard Deviation, median, max, min, etc.
  Questions:
  a. How to know which feature selection I need to use and will get the correct results?
  
  4. Image Classification: SVM
  I will try to do this step by my own after done step 1-3.
  
  
  Level Snake Segmentation (Question: What is this method use for and under which processing flow?)
  
  
  
  Thank you for your time and willing to share your knowledge 😀
• 

  Rupam Das

  1. Database: For mammogram, you need a database of atleast 50 images each of Normal and abnormal. if possible, go for three class, malignant, normal and Benign. You can find these .pgm files in MIS database. 20 is just very low number for Image Processing.
  
  2. Manual cropping is just not correct, with mamography point of view. Most suitable will be segmentation using K-means clustering. That will remove the extra part like "white written texts" from the image. No morphology processes binary images. You need the segmentation at the gray scale level.
  
  3. STFT( short term fourier transform) is theoretically best suited for non stationary signals like ECG. any tumor is best represented by change in texture. So you will be better off using either Wavelet, GLCM or even Fourier transform. Though for a change STFT can be tried out.
• 

  Rupam Das

  Mammogram Segmentation Code
  
  

  Rupam Das

  1. Database: For mammogram, you need a database of atleast 50 images each of Normal and abnormal. if possible, go for three class, malignant, normal and Benign. You can find these .pgm files in MIS database. 20 is just very low number for Image Processing.
  
  2. Manual cropping is just not correct, with mamography point of view. Most suitable will be segmentation using K-means clustering. That will remove the extra part like "white written texts" from the image. No morphology processes binary images. You need the segmentation at the gray scale level.
  
  3. STFT( short term fourier transform) is theoretically best suited for non stationary signals like ECG. any tumor is best represented by change in texture. So you will be better off using either Wavelet, GLCM or even Fourier transform. Though for a change STFT can be tried out.

  you can find the segmentation code in following address
  #-Link-Snipped-#
• 

  sharmila purushothaman

  i want matlab code for segmentation of mammogram using haar transform..