Atopic Dermatitis (AD), a chronic inflammatory skin disease causing lesions, often causes decreased quality of life (Kapur, 2018). Segmentation, a method of illustrating the difference between lesioned and non-lesioned areas of interest (AOIs) has been the primary method for which AD has been studied (Ranteke & Jain, 2013). Manual segmentation is prone to subjectivity (Ning et al., 2014) and automatic segmentation, while reliable and efficient, poses challenges such as light reflections and color variations (Lu et al., 2013). Yet, AD can be classified from color and texture (Hanifin et al., 2001; Nisar et al., 2013), as well as through machine learning methods. The purpose of this study was to determine the optimal method for segmentation of images of atopic dermatitis on subject arms in a novel and standardized photography lightbox (Lightbox) and of images of subjects' self-acquired at-home photos. The goals of this study were to determine the accuracy and reliability of photo acquisition of arms of subjects with AD in a novel standardized photography lightbox, compared to photo acquisition by subjects at home, and determine the accuracy and reliability of automated segmentation of AD lesions with combined color-based segmentation and the U-Net CNN.

No description available.

No description available.

Tato magisterská práce se zabývá využitím konvolučních neuronových sítí pro segmentační účely v oblasti transmisní elektronové mikroskopie. Také popisuje zvolenou topologii neuronové sítě - U-NET, použíté augmentační techniky a programové prostředí. Firma Thermo Fisher Scientific (dříve FEI Czech Republic s.r.o) poskytla obrazová data pro účely této práce. Získané segmentační výsledky jsou prezentovány ve formě křivek (ROC, PRC) a ve formě numerických hodnot (ARI, DSC, Chybová matice). Zvolená UNET topologie dosáhla excelentních výsledků v oblasti pixelové segmentace. S největší pravděpodobností, budou tyto výsledky sloužit jako odrazový můstek pro interní firemní výzkum.

This thesis deals with the segmentation of cranial bone in CT patient’s data after craniectomy. The U-Net architecture in 2D and 3D variant were selected for the intention of solving this problem. Jaccard index for 2D U-Net was evaluate as 89,4 % and for 3D U-Net it was 67,1 %. In the area after surgical intervention evaluating index has smaller difference between both variant, the average success rate of skull classification was 98,4 % for 2D U-Net and 97,0 % for 3D U-Net.

The aim of this thesis is segmentation of P waves in ECG signals. The theoretical part of the thesis describes the physiology of the heart and the basics of deep learning methods. Preprocessing of the signals is performed and neural network U-Net is implemented in the Python software environment in the practical part. Afterwards, optimization of network architecture is performed in order to reduce model complexity. Lastly the success rate of the model is evaluated.

This project studies the removal of non-anatomical objects from medical images. During tumor ablation procedures, the ablation probes appear in the image, hindering the performance of segmentation, registration, and dose computation algorithms. These algorithms can also be affected by artifacts and noise generated by body implants. Image inpainting methods allow the completion of the missing or distorted regions, generating realistic structures coherent with the rest of the image. During the last decade, the study of image inpainting methods has accelerated due to advances in deep learning and the increase in the consumption of multimedia content. Models applying generative adversarial networks have excelled at the task of image synthesis. However, there has not been much study done on medical image inpainting. In this project, a new inpainting method is proposed for recovering missing information from medical images. This method consists of a two-stage model, where a coarse network is followed by a refinement network, both of which are U-Nets. The refinement network is trained together with a discriminator, providing adversarial learning. The model is trained on a dataset of CT images of the liver and, in order the mimic the areas where information is missing, regular and irregular shaped masks are applied. The trained models are compared both quantitatively and qualitatively. Due to the lack of standards and accurate metrics in image inpainting tasks, results cannot be easily compared to current approaches. However, qualitative analysis of the inpainted images shows promising results. In addition, this project identifies the Frechet Inception Distance as a more valid metric than older metrics commonly used for evaluation of image inpainting models. In conclusion, this project provides an inpainting model for medical images, which could be used during tumor ablation procedures and for noise and artifact elimination. Future research could include implementing a 3D model to provide more coherent results for inpainting patients - a stack of images - instead of single images. / I detta projekt undersöks metoder för avlägsnande av icke-anatomiska föremål från medicinska bilder. Bilder tagna under ablationsbehandling av tumörer innehåller själva ablationsnålen, denna kan hindra segmenterings-, registrerings-och dosberäknings-algoritmer för att uppnå önska resultat. Dessa algoritmer kan också påverkas av artefakter och brus som genereras av olika metallimplantat. Bildifyllningsmetoder gör det möjligt att ersätta regioner som saknar eller innehåller inkorrekt bilddata, med realistiska strukturer som är sammanhängande med resten av bilden. Under det senaste decenniet har intresset för metoder för bildifyllning accelererat på grund av framsteg inom djupinlärning och ökad konsumtion av multimediainnehåll. Modeller som använder generative adversarial networks har utmärkt sig i bildsynteseringsuppgifter. Det har dock inte gjorts så många studier gällande bildifyllning av medicinska bilder. I detta projekt föreslås en ny bildifyllningsmetod för att återställa regioner med inkorrekt information i medicinska bilder. Denna metod består av ett tvåstegsnätverk, där ett första nätverk följs av ett förfiningsnätverk, båda av typen U-net. Förfiningsnätverk tränas tillsammans med ett diskriminatornätverk. Modellen tränas på ett dataset av CT-bilder av levern. För att efterlikna de områden där information saknas, applicerades masker av olika former. De färdigtränade modellerna jämfördes både kvantitativt och kvalitativt. På grund av bristen på standarder och noggranna mätvärden för bildifyllningsmetoder, kan resultaten inte enkelt jämföras med existerande metoder. Men kvalitativ analys av de målade bilderna visar ganska lovande resultat. Modellen presterar som bäst i områden inte innehåller komplexa strukturer. Sammanfattningsvis har en fungerande bildifyllningsmetod för medicinska bilder skapats och som kan användas vid tumörablation och för eliminering av bildartefakter. Framtida forskning kan inkludera implementering av en 3D-modell för att ge mer sammanhängande resultat.

In thesis, we propose a structured light vision system equipped with multi-cameras and multi-laser emitters for object height measurement or 3D reconstruction. The proposed method offers a better accuracy performance over a single camera system. The structured light method may fail the interference of reflection and scattering of light. We use U-Net to extract the laser region, obtain the laser stripe center after erosion and dilation, and finally reconstruct the point cloud corresponding to the laser stripe. Our experiments demonstrate that our structured light system with the U-Net can perform effectively and robustly in a complex environment.

Machine learning has shown great potential as a step in automating radiotherapy treatment planning. It can be used for dose prediction and a popular deep learning architecture for this purpose is the U-net. Since it was proposed in 2015, several modifications and extensions have been proposed in the literature. In this study, three promising modifications are reviewed and implemented for dose prediction on a prostate cancer data set and compared with a 3D U-net as a baseline. The tested modifications are residual blocks, densely connected layers and attention gates. The different models are compared in terms of voxel error, conformity, homogeneity, dose spillage and clinical goals. The results show that the performance was similar in many aspects for the models. The residual blocks model performed similar or better than the baseline in almost all evaluations. The attention gates model performed very similar to the baseline and the densely connected layers were uneven in the results, often with low dose values in comparison to the baseline. The study also shows the importance of consistent ground truth data and how inconsistencies affect metrics such as isodose Dice score and Hausdorff distance.

The task of upscaling pictures is very ill-posed since it requires the creation of novel data. Any algorithm or model trying to perform this task will have to interpolate and guess the missing pixels in the pictures. Classical algorithms usually result in blurred or pixelated interpolations, especially visible around sharp edges. The reason it could be considered a good idea to use neural networks to upscale pictures is because they can infer context when upsampling different parts of an image. In this report, a special deep learning structure called U-Net is trained on reconstructing high-resolution images from the Div2k dataset. Multiple loss functions are tested and a combination of a GAN-based loss function, simple pixel loss and also a Sobel-based edge loss was used to get the best results. The proposed model scored a PSNR score of 33.11dB compared to Lanczos 30.23dB, one of the best classical algorithms, on the validation dataset.