Brain tumor detection in CT data

Hello all!

I'm looking for 2d matlab implementation of random tumor detection algorithm in computed tomography images. And then should be performed a quantitative assessment of the proposed algorithm, based on the relative number of correct detections, false and invalid such discoveries.

I really need this help.

Big thanks in advance.

Regards!

Answer 1

[Matt Kindig]

Hi Peter,

Nobody is going to be able to give you a full turnkey solution to your problem, largely because a) such a task is pretty substantial and outside the scope of this forum, and b) none of us have your CT images, nor do we know what type of criteria you want to use to categorize the tumors. Do you have the Image Processing Toolbox? If so, you should start by reviewing the demos there-- several of them are relevant to image segmentation and should get you started. Pay particular attention to the use of the regionprops() function.

After you have done that, you can return here with _specific _questions. It would also be helpful if you post some sample images here to illustrate what features you are looking for.

Answer 2

[Walter Roberson]

Is it the tumor locations that are random, or is it that the detection algorithm should use randomness to pick the "locations" it will say are tumors, or is it that you have a number of detection algorithms but on any one run, some particular algorithm out of the set should be chosen randomly ?

[Peter Petrov]

Hello, Walter Roberson!

The case in point is random locations of tumor... Which is the most easily realizable detection algorithm in this case?

[Peter Petrov]

Is the following block diagram good for CT images?

MRI Scan -> [Preprocessing] -> Grayscale Image -> High pass filter -> Enhanced image -> [Post Processing] -> Threshold Segmentation -> Watershed Segmentation -> Morphological Operators (eroding, dilating) -> Output image with tumor

[Walter Roberson]

If the detection algorithm proposes a tumor position that is not exactly the same shape and position as the "known" tumor, then what measure should be used to decide how "good" the detection was?

What are the relative penalties for detecting as tumorous a pixel that is not part of the set, compared to the penalties for detecting as non-tumorous a pixel that is part of the set? Or do the penalties vary positionally? e.g., low penalty for giving an extra boundary pixel but high penalty for picking out as tumorous a pixel that is distant from any of the tumors?

What penalty for detecting an area as being two distinct tumors that touch and together cover the area that is really only one tumor?