features extraction from medical image

48 ビュー (過去 30 日間)
Majid Al-Sirafi
Majid Al-Sirafi 2017 年 10 月 12 日
コメント済み: Walter Roberson 2019 年 9 月 20 日
Hi every one
I need Matlab program for extracting features from medical image regards,
Majid
  3 件のコメント
Walter Roberson
Walter Roberson 2017 年 10 月 12 日
The code for automatically extracting any kind of features from any kind of medical image is the kind of project that you start as the topic of your PhD thesis, and then develop over the next 30 years.

サインインしてコメントする。

採用された回答

Image Analyst
Image Analyst 2017 年 10 月 12 日
  5 件のコメント
Walter Roberson
Walter Roberson 2019 年 9 月 20 日
I was part of an international multi-disciplinary brain stroke and tumor research project that included neurosurgeons.
Broadly speaking, our findings included:
  • MRI could help detect some kinds of tumors, but because of the very high importance of false negatives / false positives on brain tumors, feature detection was not a sufficient technique to be worth the effort. (Brain MRI are expensive; if you are going to do one, make sure it is worth doing; get a neurosurgeon to look at the results.)
  • CT studies were not any better (though that might perhaps be because we had experts designing new MRI pulse sequences, so we had more flexibility on MRI)
  • FMRI oxygen studies combined with image registration against the Human Brain Atlas were considerably more helpful at detecting areas affected by stroke (low oxygen use), and detecting areas where tumors were growing by finding areas consuming a lot of oxygen. Tumors detected by oxygen consumption often did not differ statistically from healthy brain images taken from MRI, and a fair portion of the time had irregular sprawling reaches of blood vessels co-opted by tumors. The somewhat dead areas cut off by stroke could be found by comparing expected oxygen use using imaging techniques, but feature shape detection on tumors detected by oxygen consumption were often a waste of time because they were so irregular; the only useful measure being difference in oxygen flow for an area (no shape detection needed.)
  • For tumors that were not obvious through FMRI oxygen study, then what really helped was MRS. We found important brain chemical differences for tumors that were detectable a fair while before MRI or CT could detect the tumor through imaging. Indeed, we found a class of brain tumor that was quite reliably, easily and comparatively inexpensively detectable through MRS of urine samples.
Thus, I do not especially recommend projects on image feature extraction to find brain tumors.
... But if you are going to go ahead anyhow, then here are some important steps to take:
  1. Do at least a week of literature research to find all of the known types of brain tumors including all of their important variants
  2. Work for at least a year on arranging to get imaging samples and tissue samples of all of those types of tumors. This will be a lot harder than it sounds, as you will need to negotiate privacy laws internationally, and you will need to negotiate patent rights, and you will need to negotiate for funding for expenses occurred in getting you the samples -- and you will need to negotiate authorship for your paper (you will not likely be successful in your negotiations unless you are writing a paper that has the backing of an internationally recognized university or an internationally recognized scientist.)
  3. Work with at least 3 neuroscientists to classify all of the parts of all of the images after microscopic examination of the tissue samples
  4. And then you can start the work on the code for feature extraction to find brain tumors.

サインインしてコメントする。

その他の回答 (0 件)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by