Nowadays surgery simulations are aiming to apply not just visual effects but forcefeedback as well. To carry out force feedback, haptic devices are utilized that are mostlycommercial products for general purposes. Some of the haptic device features are moreimportant than others in case of surgery simulator use. The precision of the output forcemagnitude is one such property. The specifications provided by haptic devicemanufacturers are lacking details on device characteristics, known to cause difficulties inplanning of accurate surgery simulations.This project shows the design of a testbed that is capable of measuring the precision ofoutput forces within the haptic devices’ workspace. With the testbed, a set ofmeasurements can be run on different haptic devices, giving as a result a betterknowledge of the utilized device. This knowledge aids the design of more precise andrealistic surgery simulations.

The purpose of this project was to analyze the coordination between gaze and the upper limb movement while performing a predefined task. We implemented a method to simultaneously compute and visualize recorded gaze data from a head mounted eye tracker and motion data from a motion capture system in the same coordinate system.  A python script was implemented to temporarily synchronize the two systems and then proceed with the spatial/coordinate transformation which was validated with the data acquired while the subject was asked to perform specific tasks. Task 1 was to fixate his gaze on a block placed in the center of a table and the Task 2 was to stack the blocks by picking it up and placing it on top of each other. Wrist and elbow flexion-extension angles were tracked simultaneously based on reflective markers trajectories while performing the task. This was visualized and discussed on how the results of our study suggest that the eye movements play a vital role in planning, estimating, coordinating and providing feedback for the body to perform a motor task.

Artificial intelligence (AI) is the study of science, engineering, and the development of intelligent machines. AI is based on human intelligence with the exception that it is not restricted by biologically observable limitations. AI has developed rapidly over the past few years and has become important all over the world. This Master’s thesis brings up AI as a medical device and the European market. The thesis provides guidance in the form of important aspects to be considered by small and medium-sized enterprises (SMEs) when marketing products in Europe. There is a lack of guidance and clear descriptions regarding AI/ML-based medical devices in Europe. Both MDR and medical devices with AI/ML are relatively new and uncharted. There are no clear guidelines, instructions, or articles that clearly describe what is needed to get an AI/ML-based medical device on the European market. In summary, there is no guidance that SMEs could benefit from when it comes to AI/ML-based medical devices and the European market. With this thesis the subject is enlightened and hopefully, the gap in knowledge about this is reduced. The chosen method to achieve the goal of this thesis is both a literature review and qualitative research in the form of interviews with relevant experts within the field. The results show that there is a lack of guidelines and regulations for AI-based medical devices, it is harder for SMEs to market such devices and it is complicated to put an AI-based medical device on the European market due to MDR. SMEs should consider certain aspects important when developing an AI/ML-based medical device and placing it on the European market. The identified aspects are creating a regulatory plan, using guidelines from example FDA, procuring regulatory competence from the start, risk classification, economics, clinical evaluation, risk management, having end-user in mind during the development, and data management/cybersecurity. The results also show that if guidelines are developed, they should contain thresholds for different characteristics in AI/ML-based medical devices, risk classification of the device, classification requirements, checklists, templates, actions, good manufacturing process description, data management, cybersecurity, patient safety process description, clinical evaluation process description, regional regulatory adaptions, and risk mitigation. The results of this thesis can be used in many ways and by many. By solely using this report for AI/ML-based medical devices, complete compliance with MDR is not fulfilled.

Atypical Femur Fracture(AFF) is a type of stress fracture that occurs in conjunction with prolonged bisphosphonate treatment. In practice, AFF is very rarely identified from Normal Femur Fracture(NFF) correctly on the first diagnostic X-ray examination. This project aims at developing an algorithm based on deep neural networks to assist clinicians with the diagnosis of atypical femurfracture. Two diagnostic pipelines were constructed using the Convolutional Neural Network (CNN) as the core classifier. One is a fully automatic pipeline, where the X-rays image is directly input into the network with only standardized pre-processing steps. Another interactive pipeline requires the user to re-orient the femur bones above the fractures to a vertical position and move the fracture line to the image center, before the repositioned image is sent to the CNNs. Three most popular CNNs architectures, namely VGG19, InceptionV3 and ResNet50,were tested for classifying the images to either AFF or NFF. Transfer learning technique was used to pre-train these networks using images form ImageNet. The diagnosis accuracy was evaluated using 5-fold cross-validation. With the fully automatic diagnosis pipeline, we achieved diagnosis accuracy of 82.7%, 89.4%, 90.5%, with VGG19, InceptionV3 and ResNet50, respectively. With the interactive diagnostic pipeline, the diagnosis accuracy was improved to 92.2%, 93.4% and 94.4%, respectively. To further validate the results, class activation mapping is used for indicating the discriminative image regions that the neural networks learn to identify a certain class.

Ionizing radiation is often used within medicine for diagnosis and treatments. Because ionizingradiation can be harmful to the body, it is important to know how it affects the tissue. Dosimetryis the study of how ionizing radiation deposits energy in a material. To measure how much ionizingradiation is deposited in the body, gas-filled detectors are often used. An ionization chamber isa type of gas-filled detector and exists in different shapes and sizes, depending on what kind ofmeasurements it is made for. Because ionization chambers are relatively expensive, it is often notpossible to buy one for each type of measurement that is to be done. This results in ionizationchambers being used for measurements they are not optimized for. This report evaluates thepossibility of 3D printing ionization chambers to make it easier to optimize them for specificmeasurements. The process included creating models of ionization chambers using CAD-software,slicing them and then 3D printing them. The 3D printed models were then brought to the SwedishRadiation Safety Authority for measurements. The ionization chambers were connected to highvoltage, and exposed to ionizing radiation in the form of high-intensity gamma-ray fields. Theoutput current of the ionization chamber was measured, which is proportional to the field intensity.The results were similar to those of a commercial ionization chamber. The conclusion is that it ispossible to 3D print ionization chambers. However, to get more accurate results, the design has tobe further optimized and more measurements need to be done.

The idea to use finite element (FE) models to reconstruct accidents for humans is becoming more popular the last years. They represent the human body very accurately and indicate well changes in shape, size and biomechanical properties. FE models are useful when looking at complex factors in the human body in a more systematic way and when the approach is too complicated for conventional setups. Positioned child models from PIPER were used in the process and then rotated in LS-PrePost according to impact points and impact velocities from a given literature data where information from witnessed fall accidents of children was given. The simulations were finally run in LS-Dyna and the purpose was to investigate if the resulting brain injuries were similar to the real life data. From the literature, the falling distance from lowest point of the body to the ground, the age of the child, gender, type of ground and results from CT scans were all known. To compare the results to the literature data, section cuts of the brain were taken at four locations with different time steps. Biomechanical injury predictors such as brain strain, acceleration, rotational angular acceleration and rotational angular velocity were observed and helped with the comparison. In total, 12 cases were reconstructed which ended as 22 simulations. Due to uncertainty regarding the falling height when the children fell from a swing, each swing case had 3 scenarios. Overall the comparison of predicted injury locations from LS-Dyna to real injury locations from CT scans indicated that 7 out of 12 cases compared relatively well. The comparison of a 23-month-old girl to the same case reconsructed with CRABI-18 showed similar outcomes of the angular acceleration and the angular velocity. The linear acceleration and HIC were however much higher with LS-Dyna. Comparison between the swing cases of a 10-, 12- and 13-year-old resulted in similar results for the 12- and 13 year-old girls but the 10 year boy had lower values for all biomechanical parameters except the angular velocity which was a bit higher. With more detailed information about real accidents and precise scaling of PIPER child models, reconstruction with LS-Dyna could be useful in the future to design safer playgrounds for children and to obtain injury criterion for children after fall incidents. / Användande av finita element (FE) modeller för att rekonstruera olyckor har blivit allt populärare de senaste åren. De representerar människokroppen mycket noggrant och indikerar väl förändringari form, storlek och biomekaniska egenskaper. FE-modeller är användbara när man tittar på komplexa faktorer i människokroppen på ett mer systematiskt sätt och när tillvägagångssättet är för komplicerat för konventionella metoder. PIPER barnmodellerna positionerades i enlighet med islagpunkter och islaghastigheter från en given databas där informationen från vittnade fallolyckor av barn gavs. Simuleringarna kördes slutligen i LS-Dyna och syftet var att undersöka om predikteringarna liknade de resulterande hjärnskadorna. Från databasen var fallhöjd från kroppens lägsta punkt till marken, barnets ålder, kön, typ av mark och resultat från CT skanningar kända. För att jämföra resultaten med litteraturdata togs sektionsavsnitt av hjärnan på fyra platser med olika tidspunkter. Biomekaniska skadeprediktorer såsom hjärntöjning, acceleration, vinkelacceleration och vinkelhastighet extraherades och användes i jämförelsen. Totalt, rekonstruerades 12 fallolyckor med totalt 22 simuleringar. På grund av osäkerhet om fallhöjden när barnen föll från en gunga, hade varje fall från gunga 3 scenarier/fallhöjder var. Sammantaget indikerar jämförelsen av förväntade skadepredikteringar från LS-Dyna till observerade skador från CT-skanningar att 7 av 12 fall korrelerade relativt bra. Jämförelsen av en 23 månader gammal tjej i samma fall som tidigare också rekonstruerades med en CRABI-18 docka visade liknande resultat av vinkelaccelerationen och vinkelhastigheten. Linjär acceleration och HIC var emellertid mycket högre med LS-Dyna simuleringarna. Jämförelse mellan fallen från gunga hos en 10-, 12-och 13-åring resulterade i liknande resultat för 12- och 13-åriga flickor, medan 10-åringen hade lägre värden för alla biomekaniska parametrar utom den vinkelhastighet som var lite högre. Med mer detaljerad information om verkliga olyckor och exakt uppskalning av PIPER barnmodeller kan rekonstruktion med LS-Dyna vara användbar i framtiden för att utforma säkrare lekplatser för barn och för att få skadeskala för barn efter fallhändelser.

Ultrasound-mediated drug delivery has been proposed as a safe and non-invasive method to achieve localized drug release. Drug-loaded microbubbles are injected in the vascular system and ultrasound waves are then used to localize and burst the microbubbles at a specific targeted area. The relatively large size of microbubbles however limits both their lifetime and their reach in the human body. Phase-change liquid droplets can extend the use of ultrasound contrast agents for localized drug delivery. Their smaller size provides several advantages. The droplets can reach smaller capillaries, such as those in tumors vasculature. Their lifetime is also considerably prolonged. Through the phenomenon of Acoustic Droplet Vaporization (ADV), triggered by ultrasound stimulation, the liquid-filled droplets experience a phase change and are converted into gas-filled microbubbles. The newly created microbubbles can then be disrupted by further stimulation and release their drug load in the tumor tissue. In this project, a protocol to image and burst perfluoropentane-based micro-sized droplets using a single transducer is developed using the Verasonics Ultrasound System. The pulse sequences are developed to allow close monitoring of the drug delivery by capturing a series of images before and after the vaporization or destruction of the droplets. The droplets response was assessed for different pulse voltages and durations. Mean pixel value was calculated for the regions of interest, using the images captured before and after delivery of the ultrasound pulse. Vaporization of the droplets can be achieved with low voltage (10V), whereas high voltage (50V) triggers their destruction. Combined with high voltage, pulse duration affects the rate at which droplets can be destructed.

The use of peruorocarbon lled droplets for use as Phase Changing Contrast Agents (PCCAs) is a promosing eld. These capsules also have potential to be used for mediated drug delivery. The phase change, which has given the capsules their name, is the process when the capsule transforms from a droplet into a bubble. This process is referred to as Acoustic Droplet Vaporization (ADV) and can be induced with the use of ultrasonic waves. In this study a new type of Perfluorpentane (PFC5) capsules which are stabilized with Cellulose Nano Fibers (CNF) have been evaluated for its potential as a PCCA. To investigate this potential a setup was designed in which the capsules could be exposed to ultrasound waves. Following the ultrasound exposure the capsules were visualized under a light transmission microscope. The experiments were conducted for dierent combinations of ultrasound parameters. For each combination eight volume distributions were created, in which two of them as reference cases were not exposed to ultrasound waves. Six cases with the ultrasound ring with different levels of acoustic power, resulting in peak negative pressures ranging from 0.144 to 0.291 MPa. The results showedfthat ADV could be observed when the frequency of the acoustic wave is 3.5 MHz, the pulse repetition frequency is 500 and the burst width is set to 12 cycles. The Peak Negative Pressure (PNP)-threshold for ADV is about 0.200 MPa. When the burst width is set to 8, ADV is also observed however to a lesser extent then when it is set to 12. These results indicate that the CNF-stabilized PFC5 capsules are promising droplets with a potential future as an alternative to currently used PCCAs.

Ice hockey, one of the most popular sports in the world, is a contact sport that is always associated with huge risks of traumatic brain injuries (TBIs) resulting from high-velocity impacts. Although technology in player protection equipment has advanced over the years, mild traumatic brain injuries (mTBIs) like concussion remain prevalent. Finite Element (FE) analysis presents a methodology to recreate accidents in an effort to study the effects of protective helmets and predict brain injuries. This study aimed at improving the response of an existing ice hockey helmet FE model during different impact conditions and reconstructing an ice hockey collision using FE simulations. First, the shear response of the Expanded Polypropylene (EPP) material for the helmet liner was improved by means of a single element simulation to replicate the experiments. Simulations of helmet drop tests were then performed to validate the helmet FE model. Two different designs of the helmet model were implemented, one with normal properties of the foam and the other with a softer foam. Actual cases of ice hockey accidents were then reconstructed using positioning and impact velocities as input from video analysis. As player to player collisions had not been reconstructed for ice hockey using two player models, it was decided to use two full body Human Body Models (HBMs) for the reconstruction. The biomechanical injury parameters for the accident reconstruction were plotted and compared with injury thresholds for concussion. The kinematic results achieved from the drop test simulations showed a considerable decrease in peak values for resultant accelerations, resultant rotational accelerations, and resultant rotational velocities. These results also exhibited better CORrelation and Analysis (CORA) scores than previously achieved. The biomechanical analysis of the accident reconstruction showed the strains in the brain for the concussed player to be more than the threshold for concussion, which confirms the validity of the reconstruction approach. The results of this study show an improved response of the helmet FE model under different impact conditions. They also present a methodology for ice hockey accident reconstruction using two full body HBMs.

Even with the recent advances of deep learning pushing the field of medical image analysis further than ever before, progress is still slow due to limited availability of annotated data. There are multiple reasons for this, but perhaps the most prominent one is the amount of time manual annotation of medical images takes. In this project a semi-automated algorithm is proposed, approaching the segmentation problem in a slice by slice manner utilising the prediction of a previous slice as a prior for the next. This both allows the algorithm to segment entirely new cases and gives the user the ability to correct faulty slices, propagating the correction throughout. Results on par with current state of the art is achieved within the domain of the training data. In addition to this, cases outside of the training domain can also be segmented with some accuracy, paving the way for further improvement. The strategy for training the network to utilise auxiliary input lies in the heavy online data augmentation, forcing the network to rely on the provided prior. / Trots att framstegen inom djupinlärning banar vägen för medicinsk bildanalys snabbare än någonsin så finns det ett stort problem, mängden annoterad bilddata. Det har bland annat att göra med att medicinsk bilddata tar väldigt lång tid att annotera manuellt. I detta projektet har en semi-automatisk algoritm utvecklats som tar sig an 3D-segmentering från ett 2D-perspektiv. En bildvolym segmenteras genom att en initialiseringbild annoteras manuellt och används som hjälp för att annotera närliggande bilder i volymen. Detta upprepas sedan för resterande bilder men istället för att manuellt annotera används föregående segmentering av närverket som hjälp. Detta tillåter att algoritmen både kan generalisera till helt nya fall som ej är representerade av träningsdatan, och gör även att felaktigt segmenterade bilder kan korrigeras i efterhand. Korrigeringar kommer då att propageras genom volymen genom att varje segmentering används som hjälp för nästkommande bild. Resultaten är i nivå med motsvarande helautomatiska algoritmer inom träningsdomänen. Den största fördelen gentemot dessa är möjligheten att segmentera helt nya fall. Metoden som används för att träna nätverket att förlita sig på hjälpbilder bygger på kraftig bilddistortion av bilden som ska segmenteras. Detta tvingar nätverket att ta vara på informationen i segmenteringen av föregående bild.