Sorry: It's not possible to reach RSS file http://schestowitz.com/Gallery/rss.php
Roy Schestowitz |
AttributionsPowered by RSS Merge v0.1.1 from Brad Touesnard. It relies heavily on FeedsBelow is a merging of various personal RSS feeds, namely:
OPMLThe feeds above (and beyond) are also contained in OPML files: Aggregation10/10/20249/25/20249/21/20249/20/20249/17/20249/12/20249/2/20248/29/20248/27/20248/23/20248/21/20248/14/20248/12/20248/8/20248/4/20246/27/20246/20/20241/1/1970Oxford Presentation on Appearance Model and Non-rigid Registration EvaluationAs anticipated, there has been little or no activity lately. However, I presented some of the latest development at Oxford last week. The presentation slides are finally online (in a variety of formats). <ul> <li><a href="/Research/Presentations/2006/MIAS-IRC-2006/final">HTML</a></li><li> <a href="/Research/Presentations/2006/MIAS-IRC-2006/MIAS-IRC_2006-final.pdf">PDF</a></li><li> <a href="/Research/Presentations/2006/MIAS-IRC-2006/MIAS-IRC_2006-final.swf">Flash/SWF</a></li><li> <a href="/Research/Presentations/2006/MIAS-IRC-2006/MIAS-IRC_2006-final.odp">OpenDocument Presentation</a></li><li> <a href="/Research/Presentations/2006/MIAS-IRC-2006/MIAS-IRC_2006-final.ppt">PowerPoint</a></li> </ul> Regarding progress, I handed in a draft of my thesis at the end of August. I am still waiting for feedback. more Presentations, Thesis and Halting of Technical Progress[<em>This will be personal item rather than a technical one</em>] am attending Medical Image Understanding and Analysis 2006, which begins later today. I will present my work tomorrow. There is also the possibility of a meeting at UCL next week -- one whose aim is to unify registration, modelling, and segmentation. It is part of a long on-going effort which occupies the entire inter-disciplinary research collaboration (IRC). I write up my thesis nowadays, so it is unlikely that I will make any technical progress in the next few months. I will try to post technical items as soon as something interesting emerges. more An Update on PublicationsS of last night, a revised submission was sent to be considered for a special TMI issue on validation. This time, for a change, we included time estimates for ground-truth-free assessement on non-rigid registration in 3-D. <br><br> Concurrently, a related paper was accepted as an oral presentation in Medical Image Understanding and Analysis (MIUA) 2006. I will post a link to that paper as soon as I put it online. more NRR Assessment: 3-D Extension and Other News<br><em>An appearance model built from37 brains <br>±2.5 standards deviations shown</em> <b>UR assessment framework, which can either evaluate non-rigid registration (NRR) algorithms or evaluate models of shape and intensity (appearance models), has been extended to 3-D</b>. It requires <em>no</em> ground truth. Among other recent development, perhaps the most notable one is that which involves running some these algorithms the in 3-D to obtain encouraging, bug-free output. Due to the immense scale of this problem, true assessment of NRR in 3-D requires clusters, where the program can be easily deployed. <br></br> In other news, we are expecting to have our work included in an issue of IEEE Transactions in Medical Imaging. The initial reviews were very positive and we are running further experiments to alleviate any doubts. more Entropic Measures for Model and NRR EvaluationURTHER on the subject of <a href="http://schestowitz.com/MARS/modules.php?name=News&file=article&sid=24">entropy in registration and model assessment</a>, there is finally an intent to publish a paper -- one which explains it in greater level of detail. The process involves treating non-rigid registration (NRR) results merely as correspondence. This correspondence builds an appearance model, for which an entropy can be computed. An entropy can be perceived as a measure of complexity as it corresponds to minimal message length that can <em>encapsulate</em> -- in this particular case -- a model of appearance. By computing such entropies, not only can the models be evaluated, but also the NRR algorithms from which these model were built. This essentially enables one to measure both NRR quality <em>and</em> model quality in an information-theoretic fashion. This method requires no ground-truth knowledge or manual annotation of any kind.<br><br> more Conference Paper on Non-Rigid Registration (NRR) ValidationOur paper to <a href="http://www.biomedicalimaging.org/">ISBI 2006</a> has just been accepted and it is available on-line as well: <ul> <li><a href="/Research/Papers/2006/ISBI_2006/Final/isbi_2006.pdf">PDF</a></li> <li><a href="/Research/Papers/2006/ISBI_2006/Final/HTML">HTML</a></li> </ul> <b>Abstract</b>: E compare two methods for assessing the performance of groupwise non-rigid registration algorithms. One approach, which has been described previously, utilizes a measure of overlap between data labels. Our new approach exploits the fact that, given a set of non-rigidly registered images, a generative statistical appearance model can be constructed. We observe that the quality of the model depends on the quality of the registration, and can be evaluated by comparing synthetic images sampled from the model with the original image set. We derive indices of model specificity and generalisation that can be used to assess model/registration quality. We show that both approaches detect the loss of registration as a set of correctly registered MR images of the brain is progressively perturbed. We compare the sensitivities of the different methods and show that, as well as requiring no ground truth, our new specificity measure provides the most sensitive approach to detecting misregistration. <br> more kNN and Entropy in Registration and Model Assessment<b>REVIOSULY we used model indices, which we called Generalisation and Specificity, to assess the quality of appearance models, as well as the quality of non-rigid registration</b>. We have now identified a valuable surrogate to these indices: Shannon's entropy. Some work by Hero <em>et al</em>. is encouraging the use of entropic measures to assess (dis-)similarity of graphs. This is practically used as non-rigid registration similarity measures -- somewhat reminiscent of mutual information (MI). <br><br> We intend to see if an entropic measure of clouds overlap suprasses the performance of Generalisation and Specificity. We also consider image distances that are based on K nearest neighbours (kNN) or the nearest match to a pixel intensity, a map of which is shown below. Since it takes around 20 minutes to generate each of the images below, we consider this to be highly impractical. To run just a single such model evaluation, we would need over 60,000 hours of computer power. And this is 2-D only... <br><br> Extension of our approach to 3-D is foreseen nonetheless. It will probably use the methods which require only a couple of hours of computation in 2-D. Resolution 'pyramiding' (coarse-to-fine approach) can assist in terms of speed. <br> <br>Top: original image; Bottom: nearest match to pixel of greyscale value 20, 60, 100 (left-to-right) for each of the other pixels in the image more Registration Assessment AbstractWe shall be presenting our registration assessment work this Tuesday. A camera-ready version of the accompanying 2-page abstract is available on-line: <ul> <li><a href="/Research/Papers/2005/MIAS_IRC_2005/Final/manchester_mias-irc_final.pdf">PDF</a> </li> <li><a href="/Research/Papers/2005/MIAS_IRC_2005/Final/HTML">HTML</a> </li> </ul> more Registration BenchmarkE may soon start working in collaboration with the <a href="http://www-ipg.umds.ac.uk/mias/gridprojects/miasgrid.shtml">MIAS-Grid</a>, as well as the IRC. MIAS-Grid, where MIAS stands for Medical Images and Signals, is a project which ultimately produces an e-Science workbench for medical image analysis. To demonstrate the utility of the system, a series of use-cases is required and our code might be among these. <br><br> Essentially, the Grid might have our registration assessment algorithms re-created. It will then compartmentale the processes and carry out some analysis in a transparent way that has robust, well-understood interfaces (e.g. <a href="http://www.xmlrpc.com/">XML-RPC</a>). Subsequently, these processes can be embedded as workflows within the workbench, which might involve autonomous and powerful computers. Algorithms can easily be exchanged, thus enabling benchmarks (comparisons) to be rapidly conducted. The idea is reminiscent of the principles and rationale behind the Strategy Patten in OO programming. <br><br> <a href="/IMG/research/registration_benchmark_framework.png"></a><br> Image registration assessment: the benchmark architecture<br> Click image for full-sized version <br><br> We are not too certain about the future of this initiative and, in particular, some of the technicalities. Yet, we would feel privileged to have an opportunity to work on such modern computing architectures. Our particular set of binaries can directly benefit from parallel workflows. In short, here is the framework that can be envisioned already: <ul> <li>We are given a set of N images</li> <li>We have M such image sets</li> <li>We need to build a model for each set among these M sets. That can definitely be done in parallel and there is no apparent dependency</li> <li>We now proceed to evaluating M models and again there are no dependencies among the evaluation processes</li> </ul> To add some context, <a href="http://www.schestowitz.com/MARS/modules.php?name=News&file=article&sid=16">registration is evaluated through the construction of appearance models</a>. All in all, the process in question need not be serial and it can be handled merely (not entirely) in parallel. We can further refine speeds by treating sub-sets of data (chunks) and then aggregating the results, if needed. <br><br> This would be similar to things we have done in the past, such as deploying banaries in computer clusters, invoking them via SSH, and collecting the output later. At extremity we used <a href="http://schestowitz.com/Weblog/archives/2005/02/25/a-necessary-hog/">30 units overnight</a> to produce some urgently needed results. <br><br> more Evaluating Appearance Models of the FaceHERE are some encouraging results (shown above), which show that we can evaluate face models quite reliably. We build <a href="http://schestowitz.com/Weblog/archives/2005/02/13/appearance-models/">appearance models</a> from a set of 68 facial images and control the quality of such models by distorting the effectiveness of manual markup. We then evaluate the models using a technique described previously in the <a href="http://schestowitz.com/MARS/modules.php?name=News&file=article&sid=19">context of MR brain datasets</a>. More technical and comprehensibe reports on face models evaluation: <ul> <li>Evaluating Registration - <b>Draft</b> (<a href="/Research/Papers/2005/Evaluating_Registration_Draft/evaluating_registration.pdf">PDF</a>, <a href="/Research/Papers/2005/Evaluating_Registration_Draft/HTML">HTML</a>)</li> <li>Generic Model Evaluation Method (<a href="/Research/Papers/2005/Generic_Model/generic_model_evaluation_method.pdf">PDF</a>)</li> </ul> more |
Dynamic page set up on February 25th, 2006 | Maintained by Roy Schestowitz |