Second Scientific Seminar - abstracts
Medical Data Collection WP4 – Paweł Mielnik (HF)
Progress in work package 4 was discussed. We have achieved all milestones: ethics committee approval was obtained. We have collected over 2000 ultrasound images from 44 patients. Data was sent to annotation.
Clinical Data Annotation WP5 – Marcin Fojcik (HiSF)
All milestones in Work Package are achieved. We trained annotation group, got necessary software from WP3, and described all of the received USG pictures. It took some time to clarify how and in which form all the medical data should be described. After many personal and skype meetings, in cooperation between software engineers and medical staff the prototype version of annotations software was projected, developed and tested. Due some collisions between study program (examinations session, summer holidays) and project schedule we had to train more than one annotation group. The next challenge is annotations of additional USG pictures. All pictures in database were described and verified (end of August 2014), but WP4 still are collecting pictures. It is done because the more pictures will be used to prepare software the more precise results in recognition can be achieved. Form the formal side WP5 is completed, but we are continuing annotations and verifying of incoming pictures.
Verification and Assessment of the Medical Diagnosis Platform WP6 – Paweł Mielnik (HF)
Progress and challenges in actual work war discussed. We elaborate the study protocol and other documentation required by ethics committee. Ultrasound data will be collected from patients with chronic arthritis. All pictures will be assessed independently by 3 rheumatologists and the software. Result will be blinded until all images are analyzed.
Noise Reduction in Ultrasound Images Based on Geodesic Paths – Bogdan Smołka (SUT), Krystyna Malik (SUT), Bernadetta Machała (SUT)
In this research a novel approach to the problem of speckle noise suppression in ultrasound images is presented. The described method is a modification of the bilateral denoising scheme and is based on the concept of local neighborhood exploration. The proposed filtering design, like the bilateral filter, takes into account the similarity of pixels intensities together with their spatial distance, and the filter output is calculated as a weighted average of the pixels belonging to the filtering window. The weights assigned to the pixels are determined by minimum connection costs of digital paths joining the central pixel of the filtering window and its neighbors. The comparison with existing denoising schemes shows that the new technique yields significantly better results in case of ultrasound images contaminated by multiplicative noise.
Identifying a joint in medical ultrasound images using trained classifiers - Jakub Segen (PJIIT), Kamil Wereszczyński (PJIIT), Marek Kulbacki (PJIIT)
A novel learning approach for detecting the joint in ultrasound images is proposed as a first step of an automated method of assessment of synovitis activity. The training and test data sets consist of images with labeled pixels of the joint region. Feature descriptors based on a pixel’s neighborhood, are selected among SURF, SIFT, FAST, ORB, BRISK, FREAK descriptors, and their mixtures, to define the feature vectors for a trainable pixel classifier. Multiple pixel classifiers, including k-nearest neighbor, support vector machine, and decision tree classifier,are constructed by supervised learning. The AUC measure computed from ROC curves is used as the performance criterion for evaluation. The measure is used to compare and select the best mixture of image descriptors, forming a feature vector for the classifier, the best classifier and the best chain of image preprocessing operations. The final joint detector is a result of clustering the pixels classified as ”joint”. The results of experiments using the proposed method on a set of ultrasound images are presented, demonstrating the method’s applicability and usefulness.
Structure matching - Jakub Segen (PJIIT), Kamil Wereszczyński (PJIIT), Marek Kulbacki (PJIIT)
Principle of the structure maching approach is described. A functioning program wchich implements structure maching algorithm is presented and demonstrated with graphics output.
Synovitis estimator: summary and status - Jakub Segen (PJIIT), Kamil Wereszczyński (PJIIT), Marek Kulbacki (PJIIT)
The summary of synovitis estimator is presented including the flow chart of it’s architecture with it’s major functional blocks. The status of the acomplishement of the work towards the estimator is described including estimated times of completion of the principal modules.
WP3 Prototype Software Development - Jakub Segen (PJIIT), Kamil Wereszczyński (PJIIT), Marek Kulbacki (PJIIT)
Implementation of methods and algorithms for synovitis degree detection requires creating an experimental environment. We defined a way how experiments will be executed and developed - the experimental environment “MEDUSA Script”. Using this environment we decreased an amount of work and conducted over 668 new experiments faster during joint detector selection process. We also upgraded and updated the Annotation Editor: existing bugs were fixed, stability and performance were enhanced and we developed data exporter. The works in the area of experiment automisation by the use of data flow was started. Continuous Integration system was upgraded to new IO and RT versions.
Conformity assessment – Sebastian Hein (ITAM)
This presentation includes information on the assessment of compliance of previously made software with the requirements of the Medical Device Directive. The result of this evaluation is that this part of the project is not subject to these requirements. This presentation also contains information about documents prepared for further work.
Speckle Noise ReductionTechnique BasedonNon-LocalMeansApproach - Krystian Radlak (SUT), Bogdan Smolka (SUT)