Showing 1–2 of 2 results for author: Tiirola, J
-
A Parts Based Registration Loss for Detecting Knee Joint Areas
Authors:
Juha Tiirola
Abstract:
In this paper, a parts based loss is considered for finetune registering knee joint areas. Here the parts are defined as abstract feature vectors with location and they are automatically selected from a reference image. For a test image the detected parts are encouraged to have a similar spatial configuration than the corresponding parts in the reference image.
In this paper, a parts based loss is considered for finetune registering knee joint areas. Here the parts are defined as abstract feature vectors with location and they are automatically selected from a reference image. For a test image the detected parts are encouraged to have a similar spatial configuration than the corresponding parts in the reference image.
△ Less
Submitted 30 June, 2023;
originally announced July 2023.
-
A Neural Template Matching Method to Detect Knee Joint Areas
Authors:
Juha Tiirola
Abstract:
In this paper, new methods are considered to detect knee joint areas in bilateral PA fixed flexion knee X-ray images. The methods are of template matching type where the distance criterion is based on the negative normalized cross-correlation. The manual annotations are made on only one side of a single bilateral image when the templates are selected. The best matching patch search is formulated a…
▽ More
In this paper, new methods are considered to detect knee joint areas in bilateral PA fixed flexion knee X-ray images. The methods are of template matching type where the distance criterion is based on the negative normalized cross-correlation. The manual annotations are made on only one side of a single bilateral image when the templates are selected. The best matching patch search is formulated as an unconstrained continuous domain minimization problem. For the minimization problem different optimization methods are considered. The main method of the paper is a trainable optimizer where the method is taught to take zoomed and possibly rotated patches from its input images which look like the template. In the experiments, we compare the minimum values found by different optimization methods. We also look at some test images to examine the correspondence between the minimum value and how well the knee area is localized. It seems that making annotations only to a single image enables to detect knee joint areas quite precisely.
△ Less
Submitted 23 September, 2022;
originally announced September 2022.
