I have MRI images of brain tumors collected from a hospital (not a benchmark dataset). And I am planning to use them to predict/classify tumour types using a typical machine learning approach: texture analysis for feature extraction to build a classification module. My question is, how to decide whether these images needs pre-processing or not, such as noise removal by using median filter for example? In other words, how to determine if there is noise in this data?
2 Answers
It depends on how did you collect these MRI images from hospital and how do you want to use these MRI images.
- If you want to use these data in your deep learning model for segmentation or detection, it is possible to skip preprocessing.
- If you want to apply a classical data science approach, you can do some processing like feature extraction.
- Lastly, from another perspective, if you want to use also patient history with images, preprocessing can be necessary.
-
$\begingroup$ Thank you for your reply, I am planning to use these images for classical data science approach as you mentioned, a typical supervised learning approach to classify brain tumours, using textural analysis for feature extraction and Support Vector Machine to build a classification module. By pre-processing, I meant, preparing the data before extracting features, such as noise removal. $\endgroup$– AtheerCommented Jan 20, 2019 at 11:35
-
$\begingroup$ @gin, notice that convolution networks are not much different from brute-forcing a range of filters, so definitely worth considering this path as well. $\endgroup$– maptoCommented Feb 19, 2019 at 12:07
I would suggest you try CNNs as they can save you a lot of preprocessing steps. In case you are going the classical ML way, it is hard to comment on the preprocessing of images without looking at them. If you are looking to reduce noise, you would probably want to apply smoothing and blurring. There are many blurring and morphological functions available in opencv library. You should try them out.