Demitri, Nevine (2016)
Signal and Image Processing Techniques for Image-Based Photometry With Application to Diabetes Care.
Technische Universität Darmstadt
Ph.D. Thesis, Primary publication
|
Text
kt_thesis.pdf Copyright Information: CC BY-NC-ND 4.0 International - Creative Commons, Attribution NonCommercial, NoDerivs. Download (3MB) | Preview |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Signal and Image Processing Techniques for Image-Based Photometry With Application to Diabetes Care | ||||
Language: | English | ||||
Referees: | Zoubir, Prof. Abdelhak M. ; Krim, Prof. Hamid | ||||
Date: | 2016 | ||||
Place of Publication: | Darmstadt | ||||
Date of oral examination: | 14 June 2016 | ||||
Abstract: | This PhD thesis addresses the problem of measuring blood glucose from a photometric measurement setup that requires blood samples in the nano litre-range, which is several orders of magnitude less than the state of the art. The chemical reaction between the blood sample and the reagent in this setup is observed by a camera over time. Notably, the presented framework can be generalised to any image-based photometric measurement scheme in the context of modern biosensors. In this thesis a framework is developed to measure the glucose concentration from the raw images obtained by the camera. Initially, a pre-processing scheme is presented to enhance the raw images. Moreover, a reaction onset detection algorithm is developed. This eliminates unnecessary computation during the constant phase of the chemical reaction. To detect faulty glucose measurements, methods of texture analysis are identified and employed in a detection scheme. Image segmentation constitutes an important contribution of this thesis. In this work, the segmentation of glucose images is interpreted as a clustering problem and is tackled using two approaches. In the first approach the mean-shift and the medoid-shift algorithms are adapted to the glucose segmentation problem. Moreover, two novel variations are derived: the robust mean-shift and the sparse mean-shift. The former is able to deal with heavy-tailed noise in the images. The latter reduces computation of the mean-shift algorithm, while not compromising in accuracy, by only processing a subset of the data vectors. The robust and sparse variations are applied to the medoid-shift. Convergence proofs for all developed algorithms are provided. The second approach to tackle the image segmentation problem is based on the union of subspaces model. To this end, sparse subspace clustering and low-rank representation methods are adapted to the glucose segmentation problem. Two novel formulations of sparse subspace clustering are derived that outperform the state of the art in producing more accurate segmentations of the video data. Additionally, a sequential version of subspace clustering is derived that achieves more accurate and earlier segmentations results. Convergence detection of the chemical reaction constitutes a further major contribution of the presented framework. For this purpose, a non-linear model of the chemical kinetics is established. Thereby, a linear relation between the reaction rate and the convergence value is exploited to facilitate the application of state estimation and prediction techniques. These succeed at increasing the accuracy of the converged remission value and decreasing measurement time drastically. To this end, the first technique is based on the extended Kalman filter and the second technique is a region-based particle filter. An extensive set of real glucose measurements is used to prove the validity of the developed approaches w.r.t. segmentation results, clinical accuracy and measurement time. Furthermore, a study is performed using real data, determining a lower limit for the blood sample volume. |
||||
Alternative Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-55641 | ||||
Classification DDC: | 600 Technology, medicine, applied sciences > 600 Technology 600 Technology, medicine, applied sciences > 610 Medicine and health 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering |
||||
Divisions: | 18 Department of Electrical Engineering and Information Technology 18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications > Signal Processing |
||||
Date Deposited: | 01 Sep 2016 10:19 | ||||
Last Modified: | 09 Jul 2020 01:20 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/5564 | ||||
PPN: | 38648807X | ||||
Export: |
View Item |