Noise removal from Optical Coherence Tomography (OCT) images

Saad Ullah

Noise removal from Optical Coherence Tomography (OCT) images

A Brief Background:

Optical Coherence Tomography (OCT) is a growing technique in medical imaging that helps in the detection of various eye diseases including retinal diseases and glaucoma mainly. This type of imaging is accomplished by using techniques of interferometry. Typically a Michelson’s Interferometer is used in taking a single point OCT image. In the same multi-point OCT images are calculated. Interferometry basically applies the technique of path difference and calculate it from fringes formed due path difference. Modern OCT devices form the 3D image of the retinal part of the eye. This 3D image is used by computer to compare with statistics of healthy eye.

Need for noise removal:

Noise removal is the part of pre-processing of almost every application. It is required for vanishing the excess or unwanted information before moving to the exact scenario of the particular application. The attributes of noise like high frequency, discontinuity and random amplitude are widely applied to attain noise removal.

In interferometry, the medium used can also cause to change the path of light, thus inserting a path difference that is not due to the comparison of the light rays. So fringes formed will produce a wrong illusion. Eventually this can lead to false analysis and prediction of the diseases.

As reflection is purely related with the brightness and distance factor of light, so due to unfriendly brightness conditions a noise is introduced that is known as the speckle noise. A typical example of speckle noise is the twinkling of stars at night. This is caused due to distance and medium. Hence this type of noise can result in fading the separation between the retinal layers of the eye resulting in distorted 3D spectrum.

As OCT imaging is a process of image acquisition, so noise is always there because of absence of ideal conditions in a practical scenario.

Hence it is a deliberate need for noise removal from the image. Also it is the first step to analyze an image to extract information.

Methodologies to attain scope of the project:

To attain the scope of the project different noise removal techniques are used and analyzed. The method, application and simulation results of each technique applied, are mentioned as the report goes on.

These techniques are basically mean, median and masking techniques.

The applied techniques include the following:

Contra-Harmonic Filtering:

This type of filtering has a significant role in the removal of salt and pepper noise. It is a neighbor evaluation technique that is applied by using the following formula:

 

                       

 

Where      Sxy  = Region covered by the order of the filter

               g(s,t)= Pixels of the Sxy

        Q = Order controlling element of numerator and denominator

            It is a restoration filter. For salt noise removal the value of Q is negative and for pepper noise removal, it is positive. If value o Q is 0 then this filter behaves as an Arithmetic-Mean filter.

            In the code an OCT image is read. Then converted to gray scale. Filtering is then applied to attain the results.

            Following is the gray scale OCT image:

After application of filter for Q=-1 following result was obtained:

            It is clearly noted that the salt noise is much reduced. Now following were the results for +1 as a value of Q of the gray scale image when filtered:

            It can be clearly observed that the pepper noise is removed. Below mentioned are the combined results of applying salt and pepper techniques sequentially:

            During the application it was observed that the application of pepper noise removal after the salt noise removal has better results than the reverse order of application.

Mean Filtering:

Mean filter also have a significant role in denoising the image. Mask of odd order like 3×3, 5×5, 7×7 etc., is applied to the image. Then mask is applied on each and every pixel one by one. Afterwards mask and image coefficients are multiplied and averaged out.

The step is repeated for every pixel in the whole image. The result of the image is displayed as below:

It can be observed that the noise is being averaged out as compared to the original one.

 

 

 

 

The noise reduction in the filtered image can be easily seen.

Median Filtering:

Median filter also have a significant role in de-noising the image. Mask constraint for this filter is same as the mean filter.

Then mask is applied on each and every pixel one by one. Then median of elements covered

in mask is taken and then placed in the center. This step is again repeated in the whole image.

The result of the image is displayed as below:

 

 

It can be observed that the noise is being averaged out as compared to the original one.

Adaptive Median Filtering:

       Adaptive Median Filter has better results in compare with the above three.

 

Conclusion:

Each one of the above mentioned techniques has its own significance as per requirement. These techniques can be used to average out the noise. The extended technique used for speckle or multiplicative noise removal is the wavelet technique that is accomplished by wavelet transform.

 

%d bloggers like this: