Nowadays hybrid imaging modalities are new trends in medical imaging. To improve the diagnostic outcome of hybrid imaging, multimodal contrast agents need to be developed. For example, hybrid contrast agents for computer tomography and ultrasound (CACTUS) are one of those desirable hybrid contrast agents for the modern medical imaging. Polyvinyl alcohol (PVA) micro-bubbles (MBs) are one of the latest generations of ultrasound contrast agents (UCAs). PVA MBs are more stable and offer longer circulation and on-shelf storage time compare to other UCAs. However, the current use as contrast agent is limited only to ultrasound imaging. In this project, we fabricated and characterized hybrid contrast agents based on PVA MBs. Two methods for developing hybrid contrast agents were proposed. The first method is to combine MBs, currently used as an ultrasound contrast agent, with gold nanoparticles that are used as a preclinical contrast agent for computer tomography (CT). The second method is to determine at which concentration plain MBs suspension has both considerable negative contrast in CT and enhancement of the backscattered signal in ultrasound imaging. Both methods were evaluated and optimized. A scenario to achieve promising hybrid contrast agent was described in this report.

Patients admitted to intermediate care units require frequent monitoring by hospital personnel. An automatisation of this monitoring would save a considerable amount of resources and could also improve the quality of the treatment. In this thesis, a deep learning-based video action recognition model is through different transfer learning approaches trained to distinguish between behaviours of patients in TV-series and a prediction system which collects, processes and predicts on images in real-time is proposed. The results from the model-training suggest that it is possible to detect behaviours that need human intervention but training on a large-scale, real-life dataset is required to form a solid conclusion. The performance results of the prediction system show that live-streamed predictions are possible at frame rates sufficient for capturing sought events, without GPU acceleration. / Patienter inlagda på intermediärvårdsavdelningar behöver frekvent övervakning av sjukhuspersonal. En automatisering av denna övervakning skulle spara en betydande mängd resurser och även kunna förbättra kvaliteten av behandlingen. I detta examensarbete tränas en djupinlärningsbaserad modell för videohandlingsigenkänning att, genom olika överföringsinlärningsmetoder, skilja på beteenden mellan olika patienter i TV-serier och ett prediktionssystem som insamlar, processerar och predikterar på bilder i realtid presenteras. Resultaten från modellträningen tyder på att det är möjligt att detektera beteenden som kräver mänsklig interaktion men träning på ett storskaligt, realistiskt dataset krävs för att kunna dra en säker slutsats. Prestandaresultaten från prediktionssystemet visar att live-strömmade prediktioner är möjliga vid bilduppdateringsfrekvenser tillräckliga för att fånga de sökta händelserna, utan GPU-acceleration.

Laser speckle contrast imaging (LSCI) is a non-invasive method for assessment of microcirculatory blood flow. The technique is based on analysis of speckle patterns to build 2D maps of perfusion with high spatial and temporal resolution. A drawback of the method is that it is highly sensitive to motion artifacts since the perfusion estimates are based on quantification of the motion blurring in the images. The camera is today limited to a bulky stand for good measurements, but even as it is fixed, it does not ensure that the patient is completely still. In many clinical settings, it would be advantageous to have a more flexible camera and to be able to detect if an image is influenced by external motion. Multi-exposure laser speckle contrast imaging (MELSCI) is an extension to LSCI that utilizes the contrast from multiple exposure times. The gain in information has paved way for more accurate perfusion estimates. The technique has been limited due to its computational complexity, but recently a real time system has been developed. The goals of this thesis was twofold, firstly find a quantifiable measure of motion artifacts to be able to evaluate and compare LSCI and MELSCI. Secondly, propose an algorithm that detects movements in LSCI recordings. Motion artifacts in LSCI and MELSCI were investigated by developing a setup where repeatable movements could be made. Measurements of a hand influenced by motions of different speeds and directions were acquired and the relative difference between motion and static states were calculated and compared for the two systems. The relative difference of the MELSCI measurements were lower for all speeds above 0.57 mm/s, indicating more robustness to motion artifacts. A detection algorithm using image registration to calculate the instantaneous speed in each frame of the recording was developed. The method successfully detects movements perpendicular to the camera and shows that the intensity images of an LSCI recording can be used to give a direct indication of when movement has occurred.

By taking advantage of the increasing amount of available electronic health data, applications of machine learning in the intensive care unit have the potential to improve medical diagnostics and risk stratification. This thesis proposes an approach for early onset prediction of pneumothorax with such technique, using time series data extracted from a clinical database. The prevalence of pneumothorax among patients is identified through ICD-9 codes, and labels for the onset are obtained by relying on proxies closely related to the condition. Both simple algorithms and deep learning networks are used in a sliding window-based prediction framework, and the importance of each feature is measured with weighted Shapley values. The results proved the feasibility of this approach using Long Short-Term Memory models, although the number of false positives is noticeably high. Mechanical ventilation was the most contributing feature for the outcome. In future work, the focus should be on addressing the large class imbalance that prevails, along with considering more well-founded methods of target labeling.

In the intensive care unit, patients with crucial, life-threatening conditions are admitted and need constant monitoring. Here, the need for a quick and efficient decision support tool is the greatest. The use of machine learning has shown promising results in identifying patients at risk of different severe conditions in the intensive care unit and detection at an early stage is crucial in order to take preventive measures. This especially applies to conditions that can be hard to manage once developed, such as circulatory shock. In this master’s thesis, a machine learning modeling approach is suggested to detect and differentiate the onset of three types of circulatory shock – cardiogenic, hypovolemic and septic shock. Data was used from the open-source database MIMIC which represents thousands of patients from intensive care. The data was preprocessed and labels for the three shock types were created using ICD-9 codes combined with a proxy that is closely related to the condition – vasopressor. Different machine learning algorithms were then used for a static onset prediction as a base. The best performing models were also trained for a dynamic onset prediction in order to make predictions up to four hours ahead of onset. All models were evaluated using different evaluation metrics and at last, an interpretation method was used to enable a simpler interpretation of the results. The final results show that it is possible to detect and distinguish between the three types of shock, up to four hours ahead of onset. For future developments, further development and validation using more data should be the main focus before testing it in a clinical setting.

Schemaläggning av personal kan vara tidskrävande, men detta kan underlättas med hjälp av digitala system. På flera närakuter i Stockholms län utförs schemaläggning för sjuksköterskor och undersköterskor fortfarande på papper, vilket medför administrativa svårigheter. Syftet med detta examensarbete var att utveckla en Android mobilapplikation som kan ersätta det nuvarande schemaläggningssystemet. Målet var att skapa en applikation som hanterar schemaläggningen av arbetspass och underlättar kommunikationen mellan all personal. Med hjälp av Android Studio utvecklades en mobilapplikation som tillåter användare att skapa schemaläggare- eller personalkonto. Användare med schemaläggarkonto kan skapa arbetspass, välja personal till och ta bort personal från passen. Användare med personalkonto kan göra en intresseanmälan för arbetspass och få en lista över sina kommande pass. Applikationen är försedd med en veckovis kalender som visar de skapade passen och personalen som är vald till passen. Med applikationen kan användarna kontakta varandra genom en chattfunktion. Den utvecklade applikationen, som uppfyllde uppdragsgivarens mål, testades och responsen vid testningen var positivt. Applikationen är fortfarande i utvecklingsfasen och behöver kompletteras med flera funktioner för att bli mer anpassningsbart för användarna. Framtida arbeten innefattar utveckling av en iOS-version och justering av applikationen för att användas på flera närakuter. Efter dessa justeringar kan applikationen publiceras på Google Play och App Store. / Scheduling can be a time-consuming process; however, this can be facilitated by utilizing digital systems. In several local emergency wards in Stockholm, the scheduling of nurses and assistant nurses is still performed on paper, which results in administrative complications. The purpose of this bachelor thesis was to develop an Android mobile application that can replace the current scheduling system. The aim was to create an application that handles personnel scheduling and supports communication between the personnel. Using Android Studio an Android mobile application was developed that allows users to create scheduler and personnel accounts. Users with a scheduler account can create work shifts, select personnel for and remove personnel from the shift. Users with a personnel account can sign up for work shifts and receive a list of their upcoming shifts. The application provides the user with a weekly schedule which displays the created shifts and the personnel selected for the shifts. By using the application, users can communicate with each other through a chat function. The developed application, which meets the client's requirements, was tested and the feedback from the testing was positive. The application is still in the development stages and requires the addition of several functions to make it more customizable for the users. Future work involves developing an iOS-version and adjusting the application to utilize it in several local emergency wards. After these adjustments, the application can be distributed in Google Play and App Store.

The emergency department at Södersjukhuset is one of the largest in the Nordic region. There is a large workforce with several skills at different levels, which means that the placement of the personnel becomes complex. The problem with the current placement method is that it is a manual and time consuming process. This report aims to solve this problem by creating a prototype of an automated and web-based placement system. During the development of the prototype different programming languages were used in unison to create a website with different functions. The most important functions to create were a search tool for names, save the placements in the form of a PDF-file and to be able to clear the placements.   The goals of the project were satisfied. However, several improvements can be made to the prototype to make it more valuable for the workplace. The advantage of the prototype was that it saves time for the scheduling staff. Additionally, it facilitates the work for the on-duty personnel as the placement is easily visualised for every module at the emergency department. / Södersjukhusets akutmottagning är en av Nordens största. Där finns en stor personalstyrka med olika yrkesgrupper och inom dem finns flera olika kompetensnivåer, vilket medför att placeringen av personalen blir komplex. Problemet med nuvarande placeringssystem är att det är en manuell och tidskrävande process. Målet för projektet var att lösa detta problem genom att skapa en prototyp av ett automatiserat och webbaserat placeringssystem.  Under utvecklingen av prototypen användes olika programmeringsspråk i samklang för att skapa en hemsida med olika funktioner. De viktigaste funktionerna som skapades var att kunna söka efter namn, spara placeringen i form av en PDF-fil samt att kunna rensa placeringen på information.   Målen för arbetet uppfylldes. Samtidigt finns det flera utvecklingsområden för prototypen som kan göra den mer användbar och värdefull för arbetsplatsens placering av personal. Fördelen med prototypen var att den sparar tid för bemanningspersonalen samt underlättar arbetet för personalen i tjänst då hela placeringen enklare kan visualiseras på varje modul.

Försvarsmakten gav KTH:s omgivningsfysiologiavdelning i uppdrag att studera energiåtgången vid gång med night vision goggles. För att ta reda på om rörelsemönstret påverkades av night vision goggles och därmed skulle ha inverkan på energiåtgången så samlades data in med inertial measurement units (IMU), men någon fullständig analys på datan gjordes inte. För att undersöka datan från IMU skapades en mjukvara i Python. Gångcykelns karaktäristik identifierades från datan för vinkelhastighet och acceleration från en IMU på vänster ankel. Position beräknades stegvis med dubbelintegration, på så vis analyserades rörelsen för varje individuellt steg. Träffsäkerheten för gångcyklarnas medelfrekvens med nämnda metod hamnade inom 5% mot tidigare validerad data med analys av motion capture system i inomhusförsök. Andra stegparametrar som steglängd och steghöjd antar dock helt orimliga värden. Vi tror att denna orimlighet till stor del beror på tekniska fel vid insamlingen av rådatan som orsakat dataförlust då liknande metoder med framgång använts av andra inom området. Rapporten beskriver utöver metoden för själva analysen även metoden för att skapa ett användarvänligt gränssnitt för forskare att utföra den med. / The Swedish Armed Forces asked the Division of Environmental Physiology at KTH to study the energy demand whilst walking with night vision goggles. To explore whether the patterns of movement changed and hence impacted energy use, data was gathered with inertial measurement units (IMU). However, no complete analysis of the collected data was performed. To study this data, software was created in Python. The gait cycle characteristics were identified from angular velocity and acceleration data from an IMU attached to the left ankle. Position was calculated stepwise by double integration, allowing for analysis of each individual step. Mean step frequency was calculated within 5% accuracy of earlier results validated by analysis with a motion capture system indoors. Other step parameters like stride length and height on the other hand take on completely unreasonable values. We believe this to largely be due to technical errors during the collection of the data leading to data loss, since similar methods have successfully been used by others in the field. The report also discusses the production of software with a user-friendly interface to be used by the scientists performing the final analysis.

I dagsläget använder akutmottagningen på Södersjukhuset ett manuellt tillvägagångssätt när personalen ska sätta in vitalparametrar som andningsfrekvens, syremättnad, systoliskt blodtryck, pulsfrekvens, medvetandegrad samt kroppstemperatur i patientjournalen. Som arbetsmetod kan detta leda till antingen inkorrekta värden eller korrekta värden men för fel patient. Behovet och efterfrågan av ett mer automatiserat system finns för ett bättre arbetsflöde.    Detta projekt har resulterat i skapandet av ett användargränssnitt som har syftet att underlätta arbetet på akutmottagningen. Användargränssnittet är utvecklat i C# och utseende är baserat på önskemål från Södersjukhusets personal för ett färdigt system. Användargränssnittet kommer inte att kunna implementeras i sjukhuset utan är tänkt att användas som exempel på hur sjukhuset vill att arbetsflödet ska se ut.

Finite Element (FE) head models are very convenient tools forthe study of Traumatic Brain Injuries (TBIs) but lack significantanatomical details for the investigation of morphology or age-dependantinjury mechanisms. In this context, the use of deformable registrationalgorithms for the generation of personalized head models is veryconsistent for the development of improved protection systems likehelmets. This thesis presents the performances of the registrationpipeline Demons combined to the Difformable Registration via AttributesMatching and Mutual-SaliencyWeighting (DRAMMS) for the generationof FE head models. Twelve subject-specific models are formed bymorphing the baseline mesh with the displacement fields resultingfrom the registration methods. The obtained models are assessedand compared through the evaluation of elements’ quality by analysisof the distortion index distribution. The Dice similarity coefficientis also calculated to estimate the personalization accuracy of theapplied pipeline. The Demons+DRAMMS registration pipeline showssatisfactory personalization accuracy for cranial mask and internalbrain structures. No significant degradation of mesh quality dueto the morphing process or specific subject morphology is observed.The present work corroborates previous study regarding the use ofDemons+DRAMMS registration pipeline for generating subject-specifichead models and validates the performances of the registration methodsand the repeatability of the morphing process for this purpose.