A solution to distinguish ventricular fibrillation and ventricular flutter from other arrhythmias and from disturbances caused by body motion or muscle activity with the use of a neural network has been investigated. Ventricular fibrillation and ventricular flutter occurs when the cardiac muscle cells are not triggered by the cardiac conduction system, but rather by ectopic foci preventing a synchronized contraction of the cardiac muscle cells and therefore inhibiting the hearts capability to properly pump blood. Two different methods, gradient descent and quasi-Newton, used by the network for learning was tested and preprocessing methods used on the input data before introducing it to the network was evaluated. Gradient descent makes use of the gradient to the error function with regards to its weights and updates the network in the direction which the output error by the network decreases the most. Quasi-Newton update the network roughly in the Newton direction by iteratively build up an approximation to the Hessian of the error function with the use of information from the gradient. The preprocessing methods used were: Threshold Crossing Intervals (TCI) which looks at the time between baseline crossings of the ECG signal. Mean Absolute Value (MAV) which computes the mean absolute value of the normalized ECG signal. Spectral Analysis which takes into account different properties of the frequency spectrum of ventricular fibrillation and normal sinus rhythm. VF-filter which assumes VF to be sinusoidal and computes the leakage after the ECG signal has been bandstop filtered around the mean frequency. Period and Amplitude Information of the maximum amplitude of the input frequency spectrum and its period. It was found that the networks that used the preprocessed signal was a poor classifier for the arrhythmias partially because ventricular fibrillation was not easily separable from the arrhythmias by the implementaion of the preprocessed inputs given.

Epilepsy is a neurological disorder affecting 1-2 % of the population in the world. People diagnosed with epilepsy are put at high risk of getting injured due to the unpredictable seizures caused by the disorder. Electroencephalography (EEG) in combination with machine learning can be used for prediction of an epileptic seizure. Therefore, a portable prediction device is of great interest with high emphasis for it to be user-friendly. One way to achieve this is by minimizing the number of electrodes placed on the scalp. This study examines the number of electrodes that provide sufficient information for prediction of a seizure. The highest prediction accuracy of 91 %, 97 % sensitivity and 85 % specificity was achieved with as few as 16 electrodes. Due to the limitation of the intracranial EEG recordings further testing must be performed on scalp EEG recordings to provide valid results.

Extrakorporeal strålbehandling av skelettsarkom är en variant av strålbehandling där en del av en patients skelett opereras ut från kroppen. Skelettsegmentet transporteras sedan vidare till ett annat behandlingsrum där den bestrålas inuti ett fantom m.h.a. en linjäraccelerator. Detta sker medan patienten är nedsövd. Efter bestrålningen kan skelettsegmentet opereras tillbaka in till patienten. På Karolinska Universitetssjukhuset i Solna görs denna strålbehandling med en metod som kräver relativt lång bestrålningstid. Detta beror på fantomets kubiska form. Ju närmare ett fantom är strålkällan desto mindre stråltid behövs. Vid det kubiska fantomet används två strålfält, ett framför och ett bakom fantomet. Det här betyder att ifall fantomets position förs närmare strålkällan, så måste den föras tillbaka lika mycket åt andra hållet efter första strålfältet. Detta gör att det blir opraktiskt samt att man inte vinner någon tid. Målet med detta projekt var att skapa ett fantom som kan förflyttas så nära strålkällan som möjligt för att minska så mycket stråltid som möjligt. Detta kommer i sin tur minska den totala behandlingstiden. Genom att skapa ett roterande cylinderformat fantom som inte är riktningsberoende, så kunde fantomet förflyttas 25 cm närmare (från 95 cm till 70 cm), jämfört med det kubiska fantomets avstånd till strålkällan. Cylinderfantomet var gjord av akrylplast och en rotationsanordning konstruerades för att rotera fantomet. Vinkelhastigheten på rotationsanordningen sattes till 15 varv/minut. Det kubiska och cylindriska fantomet jämfördes genom simuleringar. Det visade sig att bägges stråldosfördelning var likvärdiga. Bestrålningstiden kunde förkortas ner från 640 sekunder till 340 sekunder utan att negativt påverka dosfördelningen jämfört med tidigare metod.

Traditional deformable registration methods have achieved impressive performances but are computationally time-consuming since they have to optimize an objective function for each new pair of images. Very recently some learning-based approaches have been proposed to enable fast registration by learning to estimate the spatial transformation parameters directly from the input images. Here we present a method for 3D fast pairwise registration of brain MR images. We model the deformation function with B-splines and learn the optimal control points using a U-Net like CNN architecture. An inverse-consistency loss has been used to enforce diffeomorphicity of the deformation. The proposed algorithm does not require supervised information such as segmented labels but some can be used to help the registration process. We also implemented several strategies to account for the multi-resolution nature of the problem. The method has been evaluated on MICCAI 2012 brain MRI datasets, and evaluated on both similarity and invertibility of the computed transformation.

Conventional MRI techniques are not convenient when it comes to study cerebral microvascularization due to the length of the scans needed. A technique called Magnetic Resonance Fingerprinting (MRF) is an excellent candidate to solve this problem as it requires much shorter scan durations. It relies on the ability to simulate a large amount of MR signals coming from virtual voxels of controlled parameters. This thesis addresses this simulation aspect. Coding implements were made on a simulation tool called MRVox2D to improve its realism and flexibility. In particular, the voxel geometry generation algorithm was reworked to allow simulations in line with what can be obtained from a scanner. Having a variable vessel size within a simulated voxel is now possible and the Vessel Size Index can be computed accordingly. MRF applications were made on mice data using these implementations, showing encouraging but perfectible results. / Konventionella MR-tekniker är inte praktiska när det gäller att studera cerebral mikrovaskularisering på grund av längden på de skanningar som krävs. En teknik som kallas Magnetic Resonance Fingerprinting (MRF) är en utmärkt kandidat för att lösa detta problem eftersom den kräver mycket kortare skanningsvaraktigheter. Metoden baseras på förmågan att simulera en stor mängd MR-signaler som kommer från virtuella voxels av kontrollerade parametrar. Det här examensarbetet behandlar denna simuleringsaspekt. Kodningsredskap gjordes på ett simuleringsverktyg som heter MRVox2D för att förbättra dess realism och flexibilitet. I synnerhet omarbetades algoritmen för generering av voxelgeometri för att tillåta simuleringar i linje med vad som kan erhållas från en skanner. Att ha en variabel kärlstorlek inom en simulerad voxel är nu möjligt och Vessel Size Index kan beräknas i enlighet därmed. MRF-applikationer gjordes på mössdata med användning av dessa implementationer, vilket visade uppmuntrande men ännu inte perfekta resultat.

No description available.

No description available.

This project had its primary focus on running a campaign to recruit healthy volunteers for a test study involving the utilization of a microwave sensor and ultrasound measurements. The key system used was the Muscle Analyzer System (MAS), which consisted of the Bandstop Filter Sensor (BFS), a microwave sensor transmitting microwaves into the selected medium through transmission and a FieldFox Vector Network Analyzer which was used to transmit these microwaves. The second system utilized in the project was ultrasound imaging, which enabled the measurement of the size of the Rectus Femoris muscle in the thigh and the thickness of the fat and skin layers on the volunteer. This information served as a basis for interpreting the resonant frequency obtained from the MAS system. The goal was to compare these results to assess the muscle quality of the volunteers. A total of nine volunteers participated in the study, with data from four volunteers being suitable for follow- up data analysis and further research. The primary method used to obtain these results involved collecting measurements from the volunteers and comparing them with results from previous measurements conducted on sick patients in Maastricht. The resonant frequency observed for the volunteers in Uppsala was approximately 2.15 GHz, with the fat layer ranging from 5 to 18 mm in thickness, the skin layer measuring 2 mm thick, and the Rectus Femoris muscle having an area of 4 to 8 cm2. When these measurements were compared with the measurements from Maastricht on sick patients, a significant difference was observed. The patients' measurements showed values of approximately 1.98 GHz, even though there wasn't a substantial difference in the muscle and fat layer areas. In order to draw meaningful conclusions however, it would be necessary to conduct measurements using the same Vector Network Analyzer (VNA) for both healthy volunteers and sick patients with sarcopenia.

Hypoxic pulmonary vasoconstriction (HPV) is a physiological response to localized alveolarhypoxia that is intrinsic to the pulmonary circulation. By hypoxia-induced contractionof pulmonary arterial smooth muscle cells (PASMCs), the pulmonary capillary bloodflow is redirected to alveolar areas of high oxygen partial pressure, thus maintaining theventilation-perfusion ratio. Although the principle of HPV was recognized decades agothe underlying pathway remains elusive. The patch clamp technique, imaging and Ramanspectroscopy are methods that can be used to investigate parts of the mechanisms. Toenable measurements at controlled oxygen concentrations a gas-tight microfluidic systemwas developed. In this thesis preparatory experiments to couple the gas-tight systemto a microscope that enabled simultaneous measurements with patch clamp, imagingand Raman spectroscopy are discussed. The patch clamp technique is to be used formeasurements on the dynamics of the ion-channels in the cellular membrane as well aschanges in membrane potential as a response to hypoxia. Imaging of PASMCs is requiredto successfully apply the patch clamp technique. Further, imaging will also reveal whetherthe mechanical response of HPV has been triggered, for this purpose image analysis forestimation of optical flow can be used. Raman spectroscopy enables measurements ofbiochemical changes in redox biomarkers, cytochrome c and NADH, of the mitochondrialelectron transport chain. This thesis shows that the gas-tight microfluidic system providesoptimal control of the oxygen content, in an experimantal setting where the patch clamptechnique can be applied. Raman measurements showed significantly larger variationsin spectra compared to an open fluidic system, which is the conventional approach.However, the results showed a need for improved Raman preprocessing. For this purposea Convolutional Neural Network (CNN) was trained using synthetic spectra that providedoptimal reconstruction of the Raman signal. Finally, simultaneous imaging and Ramanspectroscopy of red blood cells were performed in a home built microscope. The resultspave the way for measurements on PASMCs.

Ovarian cancer is the most lethal gynaecological malignancy; however, ovarian lesions are very common and only around 1% are malignant. Due to the large number of cases, patients are triaged by gynaecologists having a high variability in diagnostic accuracy. The aim of this study is to train and validate deep neural networks and, by comparison to subjective expert assessment, determine their potential in the triage of patients with ovarian tumours. We used a transfer learning approach on pre-trained networks (VGG16, ResNet50, MobileNet), and a post-processing calibration to better align their confidence scores with the true certainty of their predictions. Our dataset contained 3077 transvaginal ultrasound images from 758 patients with ovarian tumours, where histological outcome from surgery or long-time follow-up (> 3 years) served as diagnostic ground truth. From our dataset, 150 cases (75 benign, 75 malignant), each containing 3 images, were held out for testing, while the remaining cases were used for training and model selection. The models were assessed bases on sensitivity, specificity, and AUC, along with their corresponding 95% confidence intervals. On the test set, our final model had a sensitivity of 96.0% (0.897–0.989), specificity of 86.7% (0.776–0.929), and AUC of 0.950 (0.906–0.985). When excluding the 12.7% (0.073–0.180) of cases most difficult to classify (based on the confidence score of the model output), our model had a sensitivity of 97.1% (0.909–0.994), specificity of 93.7% (0.856–0.978), and AUC of 0.958 (0.911–0.993). As comparison, the subjective expert assessment had a sensitivity and specificity of 96.0% and 88.0% respectively. We show that neural networks can be used to predict ovarian malignancy with high diagnostic accuracy, comparable to that of human experts, and thus have potential in the triage of patients with ovarian tumours. / Äggstockscancer har högst dödlighet bland gynekologiska cancersjukdomar. Äggstocksförändringar är dock vanligt förekommande och endast omkring 1% är maligna. På grund av den höga förekomsten görs initialt en bedömning lokalt (triage) huruvida patienten skall remitteras vidare för expertbedömning, eller om uppföljning på lokal vårdinrättning är tillräcklig. Triagen utförs av gynekologer som saknar utan expertkompetens inom äggstockscancer, och därav har stor variation i diagnostisk precision. Syftet med denna studie är att, genom jämförelse med subjektiv expertbedömning, utvärdera potentialen hos artificiella neurala nätverk för triagering av kvinnor med äggstockstumörer. Vi använde transfer learning av förtränade modeller (VGG16, ResNet50, MobileNet) och en kalibreringsmetod för bättre probabilistisk överensstämmelse mellan modellernas svar och deras underliggande diagnostiska precision. Vårt bildmaterial bestod av 3077 transvaginala ultraljudsbilder från 758 kvinnor med äggstockstumörer. Samtliga fall hade säkerställd diagnos genom resultat från operation eller långvarig uppföljning (> 3 år). Av detta material lades 150 fall (75 benigna, 75 maligna) à 3 bilder åt sidan för slutgiltig validering av modellen, medan resterande fall användes till träning och val av modell. Modellerna bedömdes baserat på sensitivitet, specificitet och AUC, ihop med deras 95\% konfidensintervall. Vid validering hade vår slutgiltiga modell en sensitivitet på 96,0% (0,897–0,989), specificitet på 86,7% (0,776–0,929), och AUC på 0,950 (0,906–0,985). Vid uteslutande av 12,7% (0,073–0,180) av de fall som var svårast att klassificera hade vår modell en sensitivitet på 97,1% (0,909–0,994), specificitet på 93,7% (0,856–0,978) och AUC på 0,958 (0,911–0,993). Som jämförelse hade den subjektiva expertbedömningen en sensitivitet och specificitet på 96,0%, respektive 88,0%. Vår studie visar att artificiella neurala nätverk kan användas för differentiering av benigna och maligna äggstockstumörer med hög diagnostisk precision, jämförbar med den hos erfarna läkare på området. Således bedömer vi att det finns potential för användning av dessa modeller för triagering av kvinnor med äggstockstumörer.