From a scientific perspective, is this acceptable to compare the ِactive contours technique with the deep learning technique for medical image segmentation?
and if its yes.
So what are the scientific benchmarks for comparison because these two techniques differ in approach?
Does it have scientific value?
Kindly correct me any concept that I used if it was not in a scientific way.
Active contours are like memory wire, they try to assume their predefined shape. One part (partial energy sum) tries to restore the original shape, another part tries to move the border of the contour onto detectable lines (how do you detect one?) on the image. The disadvantage is, you must have a notion of the object you want to detect. For example, if you want to find a tumour with an active contour, you have to use a shape that already resembles the tumour in the medical image. But what if the tumour has an unusual shape? What if a bone to detect is damaged? An area where active contours work best is technical objects, like implants on X-ray images.
Regards,
Joachim
They both differ in the term of how they approach the segmentation. In classic Active Contours Models (ACMs), detecting objects’ boundaries is by image gradients. However, this has one main limitation that it will be stuck at a local minimum.
One of the particular strength of CNN based models is that they work in an end-to-end fashion, which can extract hierarchical and multi-resolution features during the learning process. CNN based segmentation models can be classified into pixel-based or images-based approaches. The pixel-based approaches will classify each pixel into different objects as a classification problem.
The image-based approaches, such as U-Net, will make an image as input and output will be the segmentation of the input image (the size will be the same).
CNN can work with ACM top optimize the segmentation results.
I am agree with Mohamad M. Awad replay. However., I would like to say, yes you can do comparison with respect to performance for similar task and via similar database but different approaches is evident. Moreover, deep learning is not yet so reliable to be proposed for real medical databases for real applications. The reason is that, the reason behind high accuracy of deep learning application is not yet explained.
Active contours are like memory wire, they try to assume their predefined shape. One part (partial energy sum) tries to restore the original shape, another part tries to move the border of the contour onto detectable lines (how do you detect one?) on the image. The disadvantage is, you must have a notion of the object you want to detect. For example, if you want to find a tumour with an active contour, you have to use a shape that already resembles the tumour in the medical image. But what if the tumour has an unusual shape? What if a bone to detect is damaged? An area where active contours work best is technical objects, like implants on X-ray images.
Regards,
Joachim
I would like to add that Active Contours (AC) can consider prior knowledge. So, it would be interesting to compare both methods, particularly by comparing the size of the training set against the accuracy of the results. This is useful in medical imaging, when the database is not big enough to train a deep learning (DL) method.
So, it is posible that AC with prior knowledge outperforms DL with a small database (but AC probably needs more computation time, specially in 3D), but DL outperforms AC on large database. However, it always depends on the latent data structure.
Best Regards.
Thank you Dr. Larbi Messaouda and Dr. Muhammad Ali for your support
Chapter Fish Image Segmentation Using Salp Swarm Algorithm
Article Breast Cancer Segmentation From Thermal Images Based on Chao...
These publications may help you.
You can compare them in terms of their performance, I don't think you need to worry too much about the way of each one approaching the segmentation.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images