This paper presents a statistical framework which combines the registration of an atlas with the segmentation of MR images. We use an Expectation Maximization-based algorithm to find a solution within the model, which simultaneously estimates image inhomogeneities, anatomical labelmap, and a mapping from the atlas to the image space. An example of the approach is given for a brain structure-dependent affine mapping approach. The algorithm produces high quality segmentations for brain tissues as well as their substructures. We demonstrate the approach on a set of 30 brain MR images. In addition, we show that the approach performs better than similar methods which separate the registration from the segmentation problem.

A contextual lightweight arithmetic coder is proposed for lossless compression of medical imagery. Context definition uses causal data from previous symbols coded, an inexpensive yet efficient approach. To further reduce the computational cost, a binary arithmetic coder with fixed-length codewords is adopted, thus avoiding the normalization procedure common in most implementations, and the probability of each context is estimated through bitwise operations. Experimental results are provided for several medical images and compared against state-of-the-art coding techniques, yielding on average improvements between nearly 0.1 and 0.2 bps.

There is a need for quantitative CT data in radiation therapy. Currently there are only few algorithms that address this issue, for instance the commercial DirectDensity algorithm. In scientific literature, an example of such an algorithm is DIRA. DIRA is an iterative model-based reconstruction method for dual-energy CT whose goal is to determine the material composition of the patient from accurate linear attenuation coefficients. It has been implemented in a two dimensional geometry, i.e., it could process axial scans only.  There was a need to extend DIRA so that it could process projection data generated in helical scanning geometries. The newly developed algorithm (DIRA-3D) implemented (i) polyenergetic semi-parallel projection generation, (ii) mono-energetic parallel projection generation and (iii) the PI-method for image reconstruction. The computation experiments showed that the accuracies of the resulting LAC and mass fractions were comparable to the ones of the original DIRA. The results converged after 10 iterations.

This thesis investigated whether a deep learning model could learn features of Chaos,from the Chaos & Clues evaluation protocol, in a given dermatoscopic image data set. Asuccessful result could be of use in a future decision-support system for when dermatologists examine skin lesions for traces of melanoma (type of skin cancer). The chosen deep learning model (Inception V3) was trained to recognise four classesrelated to Chaos. Anonymous patient data was used, provided by the partnering companyGnosco. The data was partitioned into one or two classes depending on the symmetryproperties found in the corresponding image annotation. More than twenty differentmodel configurations was run to obtain the results in this thesis. The results indicate that the chosen model was not capable of learning features of Chaosfrom the dermatoscopical image data-set. Training the model to recognise features ofChaos resulted in an overfit system with low validation accuracy (close to 30%). The prediction target was changed to contrast the negative results from the Chaos classification. The chosen model was therefore configured to learn two classes, ’melanoma’ and’nevus’. This prediction target yielded a more positive result as the validation accuracywas close to 85%. However, the corresponding confusion matrix showed that these resultsare not trustworthy. It is inconclusive whether the negative results from the Chaos classification stem from thechosen approach or if the data set was insufficient for the task-difficulty. We propose adjustments to the data set for future work which could disclose if the outlined approach isviable or not.

Over 30 000 liver, abdomen, and thigh slices have been acquired by computed tomography for the SCAPIS and IGT study. To utilise the full potential of the large cohort and enable statistical pixel-wise body composition analysis and visualisation of associations with other biomarkers, a point-to-point correspondence between the scans is needed. For this purpose, an inter-subject image registration pipeline that combines the low-level information from CT images with high-level information from segmentation masks have been developed. It uses tissue-specific regularisation and processes images efficiently. The pipeline was used to deform 4603 CT scans of each slice into a respective common reference space in less than 30 hours. All but the ribs in the liver slices and the intra abdominal region of the abdomen were generally registered correctly. Vector and intensity magnitude errors indicating inverse consistency were on average less than 2.5 mm and 40 Hounsfield units respectively. The method may serve as a starting point for statistical pixel-wise body composition analysis, its association with non-imaging data, as well as saliency mapping analysis of the three-slice CT scans from the large SCAPIS and IGT cohorts.

Prosthetic heart valves (PHVs) are routinely used to replace defective native heart valves in patients suffering from valvular heart diseases. While PHVs are life-saving, they have limitations in performance and durability. Therefore, it is crucial to rigorously test and evaluate their designs before their implantation. PHVs are commonly examined using cardiovascular testing equipment that measures the hemodynamic characteristics of the valves, while also providing the opportunity for their visual assessment by collecting high-quality videos. Such visual data, obtained during mechanical simulations, are typically assessed by human experts, which is a tedious and error-prone task. Automatic assessment of PHVs from video data is possible, however, there are some challenges that need to be addressed. The evolution of the valve orifice area during one cardiac cycle is one of the key quality metrics for PHV visual assessment. Very fast motion of the valve’s leaflets is one of the challenges while dealing with the visual data. Nevertheless, the more important issue lies in the orifice being partly occluded by the inner side of the leaflets or inaccurately depicted due to its transparency. This issue has not been addressed in the literature. In the first part of the thesis, a novel orifice area segmentation algorithm is proposed for automatic quantitative performance analysis of PHVs, based on the leaflet free edges to accurately extract the actual orifice area. The video frames, recorded by a high-speed digital camera during in vitro simulations, are used to obtain an initial estimate of the orifice area using active contouring methods. This initial estimate is then refined to detect leaflet free edges via a curve extension scheme and considering brightness and smoothness criteria. Both of the developed algorithms are later modified for addressing challenges related to the fast motion of leaflets, automatic detection of the beginning of a cycle, and overly bright spots and narrow areas. Evaluation on several cases including three different PHVs and with different video qualities demonstrated the effectiveness of the proposed approach and adjustments in detecting valve leaflet free edges and extraction of the actual orifice area. The proposed method significantly outperforms a baseline algorithm both in terms of valve design and computer vision evaluation metrics. It can also cope with lower quality videos and is better at processing frames with a very small opening, which is a very crucial quality for determining the malfunctions related to improper closing of the valves. In the second part of the thesis, the above-mentioned segmented orifice area is used for the durability estimation of the prosthetic heart valves. More than 50% of PHVs encounter a structural failure within 15 years post-implantation mostly because of the excessive localized forces on some areas. We perform a computer vision (CV)-based analysis of the visual symmetry of valve leaflet motion and investigate its correlation with the functional symmetry of the valve. We hypothesize that an asymmetry in the valve leaflet motion will generate an asymmetry in the flow patterns, resulting in added local stress and forces on some of the leaflets. Two pair-wise leaflet symmetry scores are proposed based on diagonals of orthogonal projection matrices (DOPM) and dynamic time warping (DTW) techniques. The proposed symmetry score profiles are compared with fluid dynamic parameters (vorticity and velocity values) at the leaflet borders, obtained from valve-specific numerical simulations. Experiments on four cases including different tricuspid PHV designs yielded promising results, with DTW scores showing good coherence with respect to the simulations, which confirms our hypothesis. The established link between visual and functional symmetry opens the door for durability estimation of prosthetic heart valves using computer vision techniques. / Graduate

Malignant melanoma is the deadliest form of skin cancer. If correctly diagnosed in time, the expected five-year survival rate can increase up to 97 %. Therefore, exploring various methods for early detection can contribute with tools which can be used to improve detection of disease and finally to make sure that help is given in time. The purpose of this work was to investigate the performance and behavior of different convolutional neural network (CNN) architectures and to explore whether presegmenting clinical images would improve the prediction results on a binary classifier system. For the purposes of this paper, the two selected CNNs were Inception v3 and DenseNet201. The networks were pretrained on ImageNet and transfer learning techniques such as feature extraction and fine-tuning were used to extract the features of the training set. Batch size was varied and five-fold cross-validation was applied during training to find the optimal number of epochs for training. Evaluation was done on the ISIC test set, the PH2 dataset and a combined set of images from Karolinska University Hospital and FirstDerm, where the latter was also cropped to evaluate presegmentation. The achieved results for the ISIC test set were AUCs of 0.66 for Inception v3 and 0.71 for DenseNet201. For the PH2 test set, the AUCs were 0.82 and 0.73. The results for the Karolinska and FirstDerm set were 0.49 and 0.42. Presegmenting the latter test set resulted in AUCs of 0.58 and 0.51. In conclusion, quality of images could have a big impact on the classification performance. Batch size seems to affect the performance and could thus be an important hyperparameter to tune. Ultimately, the Inception v3 architecture seems to be less affected by different variability why selecting this architecture for a real-world clinical image application could be more suitable. However, the networks performed much worse than state of the art results in previous papers and the conclusions are based on rather inconclusive results. Therefore more research has to be done to verify the conclusions. / Malignt melanom är den dödligaste formen av hudcancer. Om en korrekt diagnos sätts tillräckligt tidigt kan den femåriga överlevnadsgraden uppgå till 97 %. Detta gör att forskningen efter metoder för tidigarelagd upptäckt kan bidra med verktyg som i sin tur kan användas till att upptäcka sjukdom och slutligen bidra till att hjälp sätts in i tid. Målet med detta arbete var att undersöka prestanda och beteende för olika faltningsbara neurala nätverk (CNN) och att undersöka ifall försegmentering av kliniska bilder kunde förbättra resultaten i ett binärt klassificeringssystem. De utvalda faltningsbara neurala nätverksarkitekturerna var Inception v3 och DenseNet201. Nätverken var förträanade på ImageNet och "Transfer-learning"-metoder som feature extraction och fine-tuning användes för att extrahera features från träningsuppsättningen. Batch size varierades och femtalig korsvalidering användes för att hitta det optimala antalet träningsepoker. Utvärderingen gjordes med bilder i testset från ISIC, PH2 och Karolinska och FirstDerm. Bilderna i den senare datamängden beskärdes för att utvärdera försegmenteringen av kliniska bilder. De uppnådda resultaten för ISIC testmängden var AUC-värden på 0.66 för Inception v3 och 0.71 för DenseNet201. För PH2 låg AUC-värdena på 0.82 respektive 0.73. Resultaten för testmängden med bilder frön Karolinska och FirstDerm var 0.40 och 0.42. Försegmenteringen av den sistnämnda testmängden gav AUC-värden på 0.58 och 0.51. Sammanfattningsvis kan bildkvalitet ha en stor inverkan på ett nätverks klassificeringsprestanda. Batch size verkar också påverka resultaten ochkan därför vara en viktig hyperparameter att stämma. Slutligen verkar Inception v3 vara mindre känslig för olika typer av variabiltet vilket görvalet av denna arkitektur mer lämplig ifall en riktig applikation ska byggas för detektion av exempelvis kliniska bilder. Det som bör understrykas i detta arbete är att resultaten var mycket sämre än det som bäst uppvisats i föregående rapporter och att slutasatserna är baserade på relativt ickeövertygande värden. Därför efterkrävs mer forskning för att styrka slutsatserna.

Magnetic resonance imaging (MRI) is a powerful imaging tool for diagnostics, which AMRA uses to segment and quantify certain anatomical regions. This thesis investigate the possibilities of using deep learning for the particular task of AMRAs segmentation, both for ordinary regions (fat and muscle regions) and injured muscles.The main approach performs muscle and fat segmentation separately, and compares results for three approaches; a full resolution approach, a down-sample approach (trained on down-sampled images) and an ensemble approach (uses voting among the 7 best networks).The results shows that deep learning segmentation is possible for the task, with satisfactory results. The down-sampled approach works best for fat segmentation, which can be related to the inconsistently over-segmented ground truth fat masks. It is therefore unnecessary with the additional resolution, which might only impair the performance. The down-sampled approach achieves better results also for muscle segmentation. Ensemble learning does in general not improve the neither the segmentation dice score nor the biomarker predictions. Injured muscles are more difficult to predict due to smaller muscles in the particular used dataset, and an increased data versatility. As a summary, deep learning shows great potential for the task. The results are overall satisfactory (mostly for a down-sampled approach), but further work needs to be done for injured muscles in order to make it clinically useful.

The effectiveness of colonoscopy depends on the quality of the inspection of the colon. There was no automated measurement method to evaluate the quality of the inspection. This thesis addresses this issue by investigating an automated post-procedure quality measurement technique and proposing a novel approach automatically deciding a percentage of stool areas in images of digitized colonoscopy video files. It involves the classification of image pixels based on their color features using a new method of planes on RGB (red, green and blue) color space. The limitation of post-procedure quality measurement is that quality measurements are available long after the procedure was done and the patient was released. A better approach is to inform any sub-optimal inspection immediately so that the endoscopist can improve the quality in real-time during the procedure. This thesis also proposes an extension to post-procedure method to detect stool, bite-block, and blood regions in real-time using color features in HSV color space. These three objects play a major role in quality measurements in colonoscopy. The proposed method partitions very large positive examples of each of these objects into a number of groups. These groups are formed by taking intersection of positive examples with a hyper plane. This hyper plane is named as 'positive plane'. 'Convex hulls' are used to model positive planes. Comparisons with traditional classifiers such as K-nearest neighbor (K-NN) and support vector machines (SVM) proves the soundness of the proposed method in terms of accuracy and speed that are critical in the targeted real-time quality measurement system.

No description available.