Fall-related injuries are common problems in elderly care in particular. These can cause brain damage and hip fractures, which in many cases can be serious. To reduce or mitigate the damage, various safety measures have been developed. One of them concerns a change in the surroundings, more specifically, the floor. At KTH within the Division of Neuronic Engineering, research has been done on how the impact of the case can be minimized and a shock absorbing floor (SAF) was developed. Diving problems with this type of flooring are that heavy medical beds sink into the floor, which means increased rolling resistance and thus long-term damage to both the floor and the medical staff. The aim of this thesis was to investigate how much rolling resistance can be minimized by building a downscaled hovering device based on hovercraft technology. The purpose was to enable the device to possibly be mounted under hospital beds and create a lifting force. To evaluate the performance, force measurements were performed on KTH SAF with different weights and with the use of a dynamometer. The results showed that the device reduced rolling resistance by up to 57.4% with additional weight. Four axial fans were used together with manual control of the speed of each fan. The selected components were made taking into account, in particular, cost, weight and dimensions and can thus also be limiting factors for this thesis. For future work, effective soundproofing is needed for this solution to be possible to be implemented in a hospital environment. Furthermore, studies needs to be carried out for a full-scale prototype to confirm that an equally large reduction in rolling resistance can be achieved.

Understanding the information encoded in the human genome and the influence of each part of the DNA sequence is a fundamental problem of our society that can be key to unveil the mechanism of common diseases. With the latest technological developments in the genomics field, many research institutes have the tools to collect massive amounts of genomic data. Nevertheless, there is a lack of tools that can be used to process and analyse these datasets in a biologically reliable and efficient manner. Many deep learning solutions have been proposed to solve current genomic tasks, but most of the times the main research interest is in the underlying biological mechanisms rather than high scores of the predictive metrics themselves. Recently, state-of-the-art in deep learning has shifted towards large transformer models, which use an attention mechanism that can be leveraged for interpretability. The main drawbacks of these large models is that they require a lot of memory space and have high inference time, which may make their use unfeasible in practical applications. In this work, we test the appropriateness of knowledge distillation to obtain more usable and equally performing models that genomic researchers can easily fine-tune to solve their scientific problems. DNABERT, a transformer model pre-trained on DNA data, is distilled following two strategies: DistilBERT and MiniLM. Four student models with different sizes are obtained and fine-tuned for promoter identification. They are evaluated in three key aspects: classification performance, usability and biological relevance of the predictions. The latter is assessed by visually inspecting the attention maps of TATA-promoter predictions, which are expected to have a peak of attention at the well-known TATA motif present in these sequences. Results show that is indeed possible to obtain significantly smaller models that are equally performant in the promoter identification task without any major differences between the two techniques tested. The smallest distilled model experiences less than 1% decrease in all performance metrics evaluated (accuracy, F1 score and Matthews Correlation Coefficient) and an increase in the inference speed by 7.3x, while only having 15% of the parameters of DNABERT. The attention maps for the student models show that they successfully learn to mimic the general understanding of the DNA that DNABERT possesses.

This thesis analyzes data on E. coli cell growth in a bioreactor to investigate the possibility of finding predictable correlations between the environmental parameters (sensor data) and the growth using machine learning. Discovering these correlations could be a first step toward optimizing the growth of cells to be used for cell therapy: an effective but very expensive treatment method for cancer. This could ultimately lead to decreased manufacturing costs and larger treatment availability. The data first underwent a thorough preprocessing to obtain useful features that were divided into batches. In addition, a few separate further processing methods were applied to the data for further analysis. Thereafter several different machine learning methods were implemented and evaluated on the data. All possible sensor combinations were then fed into the best-performing network and the mean absolute error was calculated for each combination. The results showed that the implemented machine learning models did not find predictable patterns between sensor inputs and growth, as the predictions did not follow the growth variations and the models mainly predicted the average yield. However, the possibility that the used approach would benefit from additional data should not be discarded. / Detta examensarbete analyserar data som beskriver celltillväxt av E. coli i en bioreaktor för att undersöka möjligheten att hitta samband mellan inputparametrar och tillväxt med hjälp av maskininlärning. Att upptäcka dessa samband kan vara ett första steg mot att optimera tillväxten av celler som används för cellterapi: en effektiv men väldigt dyr behandlingsmetod för cancer. Detta kan i slutändan leda till minskade tillverkningskostnader och en större tillgänglighet av behandlingen. All data genomgick först en ingående förberedande bearbetning för att erhålla användbara features som var uppdelade i batcher. Ett antal separata vidarebearbetningsmetoder tillämpades också för vidare analys. Därefter implementerades och evaluerades ett flertal olika maskininlärningsmetoder. Den bäst presterande modellen blev tränad på alla möjliga sensorkombinationer och medelabsolutfelet beräknades. Resul- taten visade att de implementerade maskininlärningsmodellerna inte hittade förutsägbara mönster mellan sensorinput och celltillväxt, då förutsägelserna inte följde tillväxtvariationerna och modellerna främst förutspådde den genomsnittliga celltillväxten. Trots resultatet bör möjligheten att ytterligare data kan gynna det använda tillvägagångssättet inte förkastas.

Mechanical properties of muscle tissue are linked to its function. Magnetic Resonance Elastography (MRE) has the capability toquantitively measure these mechanical properties of soft tissue in-vivo. However, most MRE methods assume tissue isotropy during MRE reconstruction which is not an appropriate assumption for muscle tissue. To obtain tissue mechanical properties with higher accuracy, muscle anisotropy must be considered during MRE reconstruction. Therefore, the aim of this thesis was to implement an anisotropic MRE reconstruction. The anisotropic MRE reconstruction used solves three independent viscoelastic parameters (G||, G⊥, E||), that is based on a transverse isotropic (TI) inversion. The reconstruction was validated in a phantom study by comparison with an isotropic reconstruction, and in-vivo on gastrocnemius (ankle plantar flexor) and tibialis anterior (ankledorsi flexor) of one human subject. Also, a TI phantom was created to be included in the phantom study along with a commercial isotropic phantom. However, due to poor image quality, the TI phantom was excluded from further validation. Results from the isotropic phantom did not agree with the assumption of equal shear modulus in different planes of the isotropic medium. In-vivo study showed that the anisotropic reconstruction yielded mean moduli values in the range of what the literature suggests. The unequal shear modulus and young's modulus G||≠G⊥≠E|| of anisotropy in TI material was observed, indicating that the reconstruction method was able to identify the anisotropy ofmuscle tissue. However, further analysis including more human subjects are needed to conclude the reliability of the reconstruction method. / Muskelvävnadens mekaniska egenskaper är kopplade till dess funktion. Magnetresonans-Elastografi (MRE) har förmågan att kvantitativt mätadessa mekaniska egenskaper i mjukvävnad in-vivo. De flesta MRE metoder antar dock att vävnaden är isotropisk under MRE rekonstruktionen, vilket inte är ett lämpligt antagande för muskelvävnad. För att med högre noggrannhet estimera mekaniska egenskaper i vävnad måste muskel anisotropi tas i hänsyn under MRE rekonstruktionen. Därav var syftet i denna uppsats att implementera en anisotropisk MRE rekonstruktion. Den implementerade anisotropa MRE rekonstruktionenlöser tre oberoende viskoelastiska parametrar (G||, G⊥, E||), som ärbaserad på en transversell isotropisk (TI) inversion. Rekonstruktionen validerades i en fantomstudie genom jämförelse med en isotropisk rekonstruktion samt in-vivo på gastrocnemius (ankel plantarflexion) och tibialis anterior (ankel dorsalflexion) hos en människa. Dessutom skapades en TI-fantom för att inkluderas i fantomstudien tillsammansmed en kommersiell isotropisk fantom. På grund av dålig bildkvalitet uteslöts TI-fantomen från ytterligare validering. Resultat från den isotropa fantomen överensstämde inte med antagandet om lika skjuvmodul i olika plan av det isotropa mediet. In vivo-studien visade att den anisotropa rekonstruktionen gav medelvärden för modulerna inom intervallet av vad litteraturen antyder. Olikheten av skjuvmodulen och elasticitetsmodulen för anisotropi i TI-material G||≠G⊥≠E|| observerades vilket är en indikation att rekonstruktionsmetoden kunde identifiera muskelvävnadens anisotropi. Det krävs dock ytterligare analys inklusive fler mänskliga försökspersoner för att dra slutsatsen ifall rekonstruktionsmetoden är tillförlitlig

Idag är unga kvinnor den grupp som utgör majoriteten av ätstörningsdrabbade, där anorexia nervosa, bulimia nervosa och EDNOS (eating disorder not otherwise specified) är de vanligast förekommande diagnoserna. Mando Group AB behandlar patienter med ätstörningar och har till patienternas hjälp utformat datorprogrammet Doktor Cecilia som är en virtuell behandlare till vilken patienterna kan ställa frågor. Användandet ligger enligt Mando Group AB inte på önskvärda nivåer och programmet har inte heller tidigare genomgått en utvärdering. Syftet med detta arbete var att ta fram potentiella förbättringar av programmet och att av patienterna få dessa tillsammans med programmet utvärderade. Vidare utfördes en enkätundersökning på två Mandometerkliniker för att kartlägga patienternas åsikter om programmet och de framtagna potentiella förbättringarna. Resultatet visade att patienterna var positivt inställda till implementerandet av bland annat röstsvar, tvåvägskommunikation och möjlighet till att se till programmet vanligt ställda frågor. Detta resultat användes sedan för att framställa rekommenderade implementationer som skulle kunna förbättra programmet och på sikt kunna öka dess användning. / Today, adolescent females represent the majority of people suffering from eating disorders with anorexia nervosa, bulimia nervosa and EDNOS (eating disorder not otherwise specified) being the most common diagnoses. Mando Group AB treats patients with eating disorders and has for the sake of the patients developed the computer software ‘Doctor Cecilia’ which is a virtual caretaker to whom patients may direct their queries. Mando Group AB has stated that the usage of the software has not reached the desirable level amongst their patients. Moreover, the software has not been subject to an evaluation. The purpose of this thesis has been to produce suggestions of potential improvements for ‘Doctor Cecilia’ by consulting patients in order to evaluate these improvements as well as the software in general. As such, an investigation was conducted at two Mandometer-clinics in order to clarify the opinions of the patients regarding the software itself and the potential improvements which are presented by this thesis. The results showed that the patients were in favor of the implementation of for instance voice response, two-way communication and the opportunity to see frequently asked questions. The results of this thesis have been used to produce recommended implementations that could improve the software and eventually increase its usage.

Any center planning to install a Gamma Knife radiosurgery unit has to provide for an efficient shielding of the treatment room, to protect the patient, the staff and the public, against undesired radiation. The shielding barrier design is controlled by national and international recommendations; the reference documents for gamma ray radiotherapy facilities are the National Council on Radiation Protection and Measurements (NCRP) reports 49 and 151. However, some facts highlighted in this thesis point out that NCRP methods are ill-adapted to the Gamma Knife. Spectroscopic measurements were performed around the Gamma Knife with a Germanium detector. They revealed that the radiation field contains few high energy photons, is highly anisotropic, and that the leakage level is much lower than the NCRP estimation. These observations led to the development of a new approach to determine the necessary shielding, based on the actual and directly measurable radiation field around the unit. This method would reduce the shielding oversizing induced by the unsuitability of the NCRP recommendations for the Gamma Knife.

In healthcare, organisational changes are suggested and implemented routinely as an answer to increasing patient demands and financial restrictions. Computer simulations offer a way to test these new procedures as well as increase the knowledge about complex systems. However, there are several challenges such as time and the difficulty to integrate simulations. To cope with these problems, ontologies is proposed as a foundation for simulations. Ontologies enable transfer of semantic meaning and a standardised way to describe objects to simplify integration of simulations. As a first step in this direction, this study aims to create an ontology framework by identifying archetypical concepts in hospital processes. The developed ontology framework is based on a generic process ontology. Moreover, case studies and interviews were used to identify, define and implement concepts in the software Protégé 5.1 using OWL-DL standard. Validation was performed by mapping hospital processes defined in three case studies using the ontology framework. The resulting ontology framework consists of 146 classes, 180 instances, 34 relations and 61 attributes. The validation of the ontology framework suggested that the framework was able torelate to all terms in the hospital processes presented in the case studies. Moreover, 50% of the terms were identical or synonyms to the concepts defined in the ontology framework. This is an indication that the ontology framework serves its function to model general concepts in hospital processes. In the future, the ontology framework will be used to create computer simulations of hospital processes. The ability to transfer semantic information and thus create computer simulations which are easy to integrate, will also be tested. Moreover, the ontology framework will be expanded by analysis of more hospital processes. / Organisationsförändringar blir allt vanligare i vården på grund av ökande patientkrav och finansiella restriktioner. Datorsimuleringar erbjuder ett sätt att testa nya processer och samtidigt öka kunskapen om komplexa system. Simuleringar erbjuder dock flertalet utmaningar eftersom det är tidskrävande och integrering av simuleringar är svårt. För att hantera dessa problem föreslås ontologier som en lösning för att skapa integrerbara simuleringar. Ontologier möjliggör överföring av semantisk mening och förenklar integrering genom att beskriva objekt på ett standardiserat sätt. Målet med denna masteruppsats är att utveckla ett ontologiskt ramverk för atti senare studier skapa integrerbara simuleringar av sjukhusprocesser. Det ontologiska ramverket skapas genom att identifiera arketypiska koncept i sjukhusprocesser. Det utvecklade ontologiska ramverket baserades på en generisk processontologi, fallstudier och intervjuer som användes för att identifiera och definiera koncept. Koncepten implementerades sedan med hjälp av mjukvaran Protégé 5.1 genom OWL-DL standard. Validering genomfördes med hjälp av fallstudier genom att se hur väl det ontologiska ramverket kunde kartlägga de sjukhusprocesser som fallstudierna beskrev. Det slutgiltiga ontologiska ramverket består av 146 klasser, 180 instanser, 34 relationer och 61 attribut. Valideringen visade att det ontologiska ramverket kunde relatera till alla termer i sjukhusprocesserna som presenterades i fallstudierna. Vidare var 50% av alla termer i fallstudiernasprocesser identiska eller synonymer till koncepten definierade i det ontologiska ramverket. Detta indikerar på att det ontologiska ramverket fyller sin funktion för att modellera generella koncept i sjukhusprocesser. I framtiden ska det ontologiska ramverket användas för att skapa datorsimuleringar och berikas genom ytterligare studier av sjukhusprocesser. Ramverkets förmåga att överföra semantisk mening och på sätt skapa integrerbara datorsimuleringar ska även testas.

Early assessment of tumour response has lately acquired big interest in the medical field, given the possibility to modify treatments during their delivery. Radiomics aims to quantitatively describe images in radiology by automatically extracting a large number of image features. In this context, PET/CT (Positron Emission Tomography/Computed Tomography) images are of great interest since they encode functional and anatomical information, respectively. In order to assess the patients' responses from many image features appropriate methods should be applied. Machine learning offers different procedures that can deal with this, possibly high dimensional, problem. The main objective of this work was to develop a method to classify lung cancer patients as responding or not to chemoradiation treatment, relying on repeated PET/CT images. Patients were divided in two groups, based on the type of chemoradiation treatment they underwent (sequential or concurrent radiation therapy with respect to chemotherapy), but image features were extracted using the same procedure. Support vector machines performed classification using features from the Radiomics field, mostly describing tumour texture, or from handcrafted features, which described image intensity changes as a function of tumour depth. Classification performance was described by the area under the curve (AUC) of ROC (Receiving Operator Characteristic) curves after leave-one-out cross-validation. For sequential patients, 0.98 was the best AUC obtained, while for concurrent patients 0.93 was the best one. Handcrafted features were comparable to those from Radiomics and from previous studies, as for classification results. Also, features from PET alone and CT alone were found to be suitable for the task, entailing a performance better than random.

Central venous catheterization (CVC) is performed frequently in clinical practices in Operation Rooms(OR) and Emergency Rooms(ER). CVC is performed for different reasons such as supply nutrition or medicine. For CVC ultrasound (US), diagnostic device is preferable among other medical imaging technologies. Central Venous Catheterization (CVC) under the 2D US guidance requires skills for catheter placement to avoid damaging non-targeted blood vessels. Therefore, there is a great need for improvements in the CVC field where mechanical complication has a risk of occurring during the procedure, for example misplacement of the needle. This master thesis project was performed at Kyushu University department of mechanical engineering and collaboration with Maidashi Kyushu University Hospital department of Advanced Medicine and Innovative Technology in Japan. This thesis is aimed to develop a new simple and cheap guidance system for CVC placement. The system performed on a Tablet PC and will be using already existing Ultrasound machines in Hospitals and Augmented Reality(AR). The project developed a simple AR-System for CVC placement with use of the ArUco library. The new developed system takes information from Ultrasound images and constructs a 3D-model of a vein and artery. Then the 3D-model is augmented on the patient through the tablet PC by using the ArUco library. The construction and augmentation is all performed on the tablet PC and has a small computation to complete the necessary procedures for the 3D-blood vessels. The AR-system has a simple control where the interface of the system is a simple push and action system. When a good visualization of the blood vessels and Doppler effect is shown on the ultrasound machine, the surgeon presses the screen and the image is saved. After about five images are acquired, the system builds the 3D-model and augments it on the patient using a marker. For the development of this system it implemented the waterfall method where each step was tested and checked, before moving to the next step. A full functional system was developed and tested. From the tests performed it is shown that there were limitations due to segmentation and depth perception. But the system has possibilities as an aid for CVC placement.

Mucins, the main macromolecular constituent responsible for gel-forming property in mucus, have great potential to act as new biological hydrogel for medical applications. Click chemistry reaction is an attractive tool to be applied in both bioconjugation and material science to form covalent bonds between molecules. Herein the click chemistry reaction of tetrazine-norbornene ligation was adapted to form click mucin hydrogel using purified commercial available bovine submaxillary mucin (BSM). This study included the characterization, purification and chemical modification of commercial available BSM. The flow filtration purification was chosen after investigating the effectiveness and yields of four different purification strategies. The reactivity of tetrazine and norbornene-functionalized BSM was evident from the formation of robust mucin hydrogel within minutes after mixing the two components. / Mucin, den viktigaste makromolekylära beståndsdel som ansvarar för den gelbildande egenskapen i slem, har stor potential att fungera som en ny biologisk hydrogel för medicinska tillämpningar. Klick-kemi reaktioner är attraktiva verktyg som kan användas i både biokonjugering och materialvetenskap för att bilda kovalenta bindningar mellan molekyler. I detta projekt användes renat kommersiellt köpt bovint submaxillärt mucin (BSM) i en klick-kemi reaktion för att sammanlänka tetrazin och norbornylen. Denna reaktion anpassades för att bilda en mucin hydrogel. Detta projekt inkluderade karakterisering, rening och kemisk modifiering av kommersiellt köpt BSM. Flödesfiltrering valdes som reningsmetod efter undersöking av effektivitet och utbyte av fyra olika reningsstrategier. Reaktiviteten hos tetrazin och norbornen-funktionaliserad BSM var uppenbar från bildandet av robust mucin hydrogel inom några minuter efter de två komponenterna sammanblandats.