|
Lääketieteellisen fysiikan ja tekniikan yhdistys (LFTY) Finnish Society for Medical Physics and Medical Engineering |
LFT-päivät 9.2.2017, Oulun yliopisto
Posterikilpailuun osallistuneet
1. Alikhani Iman, Oulun yliopistoECG-derived respiration estimation during daily activities
2. Elsayed Hassaan, Oulun yliopisto
Detecting the structure and disorders of articular cartilage using correlation time mapping
3. Iivanainen Joonas, Aalto-yliopisto
Active magnetic shield for optical neuromagnetic measurements
4. Juntunen Mikael, Oulun yliopisto
Inverse techniques for the reconstruction of sparse dual-energy projections
5. Kajabi Abdul, Oulun yliopisto
Assessment of meniscus with adiabatic T1p and T2p in asymptomatic subjects and patients with early oseoarthritis
6. Kankaala Erik, Oulun yliopisto
ICT in promoting the health of the elderly
7. Korhonen Miitu, Itä-Suomen yliopisto
Hoitokohteen liikekorjauksen tarkkuus eturauhasen robottisädehoidossa
8. Paksuniemi Iida, Oulun yliopisto
Päivittäisen fyysisen aktiivisuuden suhde kaatumisiin ikääntyneillä
9. Tervo Aino, Aalto-yliopisto
Noise optimization of multi-layer insulation in liquid-helium cryostat for brain imaging
10. Vanhatalo Joonas, Tampereen teknillinen yliopisto
Monitoring of postoperative myocardial ischemia and infarction with continuous ECG measurements
11. Zebarjadi Niloufar, Tampereen teknillinen yliopisto
Static and dynamic approaches for pain detection and pain intensity estimation using facial expression
Postereiden abstraktit
ECG-derived respiration estimation during daily activitiesAlikhani Iman, Oulun yliopisto
Physiological signals have been widely utilized for diagnostics, ambulatory monitoring of subjects for prevention, analysis and health care purposes. Electrocardiography (ECG) is one of the well-known biosignals widely used for diagnostics and nowadays for self-monitoring. It is known that the respiratory information is modulated in the ECG signal. Our aim in this study is to derive breathing rate, indirectly from subject's single-channel ECG signal recorded by sport heart rate monitors using signal processing techniques throughout uncontrolled context, including office, households and sport activities.
Our database consists of 67 subjects (30 female and 37 male) aged from 18 to 60 years old conducting several activity protocols, illustrated in Figure 1. Figure 2 also explains the results obtained with different methods throughout different activity protocols. It is visualized that in all three categories of experimented activities, fusion method is top-of-the-list with less error compared to other methods. There are some interesting results can be seen in Figure 2. It seems that RRI signal is more robust in office activities, while during house and sport activities, RPA and AUR lead to superior results.
ECG derived respiration (EDR) has been studied since mid 80th. However, not many studies have been found to evaluate EDR during uncontrolled measurement condition. 67 subjects were experimented during our measurement and their single channel ECG plus their reference breathing rate were recorded. Besides, we implemented nine different estimation models and run them over the collected database for different kinds of evaluation. Our methods are divided to eight spectral methods and a single time-domain method. HRV signal plus the amplitude of R peaks or RPA and area under R wave known as AUR are the basic signals extracted from ECG and potentially contain respiratory information. Basically, the acquired results notice that the more movement, the harder the estimation. In most of the protocols, it is illustrated that fusion models which combine the original spectrum of RRI, RPA and AUR signal lead to superior results compare to the single spectrum strategy. It is due to the reason that, using fusion technique, the common high frequency components of all these signals are intensified by element-wise multiplication of their spectrums. Although there might be some distracting components (e.g. movement artifacts and CLC components during sport activities with high degree of impact from floor) in the spectrum of those signals, their strength drops by the fusion.
Detecting the structure and disorders of articular cartilage using correlation time mapping
Elsayed Hassaan, Oulun yliopisto
INTRODUCTION: Osteoarthritis (OA) is the most common joint disease, especially among the middle aged and elderly people. It affects the joint structure and function. It starts with the degradation of the articular cartilage. Articular cartilage is a highly specialized connective tissue that provides the joint with frictionless, painless motion ability. Moreover, it bears compressive loads applied to the joint. Several magnetic resonance imaging (MRI) techniques can be used to assess the articular cartilage constituents. Quantitative MRI parameters can provide data regarding the whole joint structure. However, these parameters are mainly sensitive to the magnetization dynamics of water protons.The purpose of this work is to develop a new MRI contrast, correlation time (Ï„c), that is sensitive to the tissue structure and composition. The new MRI contrast characterizes articular cartilage properties at molecular level and it is related to the dynamics of water molecules in their microenvironment.
METHODS: Ï„c maps were fitted from T1Ï measurements of spin-lock frequencies = 125, 250, 500 and 1000 Hz of human (d = 6 mm, N = 14) and bovine (d = 25 mm, N =18) articular cartilage specimens. The human samples were classified into two groups: early and advanced OA. The bovine samples were divided into three groups: control, collagen digested and proteoglycans digested groups. Region of interest (ROI) was selected from the center of cartilage in each Ï„c map and mean profiles were calculated from the ROIs. Ï„c ROIs were also statistically compared with T2 and T1Ï ROIs.
RESULTS: Ï„c revealed the laminar appearance in control and PG depleted bovine specimens, which was not observable in the collagen-digested specimens. The laminar appearance was also visible in the early OA human specimens, in contrast to the advanced OA group. The mean profiles of Ï„c revealed complementary information to that in the T2 and T1Ï profiles. The Ï„c mean value in the superficial zone was significantly longer in the control group than in the collagen digested group. Ï„c values tend to increase near the cartilage-bone interface.
CONCLUSIONS: The results demonstrated that Ï„c mapping is a potential method for providing information about the articular cartilage structure and changes associated to cartilage degeneration. Ï„c is affected by the of collagen fibers, since Ï„c maps revealed the laminar structure of the cartilage. The Ï„c may be viewed as a novel MRI biomarker for the evaluation of changes in cartilage structure and composition.
Active magnetic shield for optical neuromagnetic measurements
Iivanainen Joonas, Aalto-yliopisto
Optically-pumped magnetometers (OPMs) are emerging as a promising alternative sensor for recording magnetic fields generated by the human brain. OPM measures the magnetic field by optically detecting the Larmor precession of a spin polarization that is generated into an alkali vapor by means of optical pumping. These OPM sensors can only operate at very low absolute magnetic fields and therefore efficient shielding against the ambient magnetic field is required. The standard in biomagnetism is to use a passive magnetically shielded room to obtain sufficient shielding; however, the remanent magnetic field in a typical shielded room often exceeds the limits of these new magnetometers. In the thesis, a suitable ultra-low-field environment for the use of OPM sensors in Aalto University was set up.
First, the magnetic fields inside the two- and three-layer magnetically shielded rooms of Aalto University were surveyed. Based on the measurements, a movable lightweight active compensation system for further reducing the fields was designed and built. The AC and DC shielding performances of the constructed system were then evaluated.
Without compensation, the measured remanent magnetic fields were on the order of 100 nT and 5 nT in the two- and three-layer rooms while the gradients were about 40 nT/m and 5 nT/m, respectively. Both rooms had low-frequency field drifts of approximately 1 nT in a measurement period of 200 s. The constructed compensation system was designed to consist of six computer-controlled coils, which can produce homogeneous fields and diagonal first-order gradients. In addition, the feedback loop of the system was designed so that it allows using same sensors both for compensation and brain measurement. With the constructed compensation coil set, the static field could be reduced by a factor of better than 12 in a head-sized volume. The feedback loop of the compensation system was also capable of locking the field to zero at the sensor site and could also remove the low-frequency fluctuations in the field.
The thesis showed that neuromagnetic measurements with optically-pumped magnetometers should be feasible in standard shielded rooms by utilizing the constructed compensation system.
Inverse techniques for the reconstruction of sparse dual-energy projections
Juntunen Mikael, Oulun yliopisto
Introduction: Sparse angle, dual-energy X-ray imaging enables low-cost and low-dose tomographic imaging with excellent tissue discrimination. The image noise, however, is higher in tissue discriminated projection images and robust tomographic image reconstruction techniques are needed in order to obtain sufficient reconstructions from these sparse and relatively noisy projections.
Aim: The aim of this study was to apply novel inverse techniques to reconstruct material decomposed tissue specific images (bone decomposed images). We propose that the noise in tissue specific reconstructions can be significantly reduced with the appropriate choice of reconstruction technique.
Methods: The reconstruction techniques compared in this study were least squares (LSQR) method, Tikhonov regularization and total variation (TV) regularization. The aforementioned tomographic image reconstruction techniques were applied to reconstruct material decomposed bone images of an anthropomorphic phantom. A calcification insert was positioned within the phantom and the visibility of the insert was compared for different reconstruction techniques.
Results: The use of regularizing reconstruction techniques significantly improved the contrast to noise ratio (CNR) of the insert with respect to the background. CNR for LSQR was 1.8 and 3.4 for Tikhonov and TV regularization. The use of these regularization techniques not only reduced the background noise but also visually improved the reconstruction of the insert itself (Figure 1).
Discussion: These findings propose that the choice of image reconstruction technique significantly affects the quality of dual-energy reconstructions when relatively few projections are used. More specifically, regularizing reconstruction techniques performed significantly better than the non-regularizing technique (LSQR) in noisy measurements.(e.g., follow-ups).
Assessment of meniscus with adiabatic T1p and T2p in asymptomatic subjects and patients with early oseoarthritis
Abdul Wahed Kajabi, Oulun yliopisto
It has been demonstrated that adiabatic qMRI relaxation parameters in rotating frame, namely AdT1p and AdT2p, are superior in detecting early degenerative changes in cartilage over numerous other qMRI techniques. Since both cartilage and meniscus consist of a macromolecular framework of collagen fibres and proteoglycans, AdT1p and AdT2p are also hypothesized to be more sensitive to early meniscal structural changes.
To assess meniscal structural changes in asymptomatic subjects and patients with early osteoarthritis using adiabatic AdT1p and AdT2p.
Quantitative assessment of meniscus was performed with AdT1p and AdT2p mapping using a 3 T clinical MRI system (Siemens Skyra) in 15 asymptomatic volunteers and 15 patients with mild OA (KL = 1,2). Sagittal images of the knee were acquired using preparation block consisting of a train of 0, 4, 8, 12 and 16 adiabatic fast passage (AFP) hyperbolic secant pulses (pulse duration = 6 ms) of the HSn family, here HS4; followed by FLASH readout (TR/TE = 4000/3.36 ms, matrix = 256x256, FOV = 180x180 mm2, flip angle = 15°, ?B1max = 800 Hz). For AdT2p, the AFP pulses were placed between two adiabatic half passage pulses. The cohorts were matched for sex and age. MR images of all the subjects were separately assessed by an experienced radiologist and scored using MOAKS. Mean AdT1? and AdT2? values were calculated in four different regions of interests of meniscus (Fig.1). Then the calculated mean values were used to assess (i) meniscal structural changes in volunteers and patients and (ii) meniscus lesion in subjects classified based on MOAKS independent of their symptoms and KL.
Both AdT1? and AdT2? in PHMED of patients were significantly longer than their respective asymptomatic compartments (p = 0.027, ? (relative mean difference) = 17.9% for AdT1?, p = 0.003, ? = 28.9% for AdT2?). Similarly, both AdT1? and AdT2? of PHMED with lesion were significantly longer than their relevant compartments with no lesion (p < 0.001, ? = 26.1% for AdT1?, p = 0.014, ? = 21.4% for AdT2?). The longer values of AdT1? and AdT2? for patients and subjects with lesion indicate the severity of meniscal structural alterations in PHMED.
Our findings indicate that AdT1Ï and AdT2Ï measurements could not only distinguish meniscal degeneration in subjects with clinical symptoms but could also distinguish the lesion in subjects clinically assessed by radiologists. AdT1Ï and AdT2Ï relaxation measurements have significant implication in detecting and monitoring noninvasively the early structural changes in meniscus associated with meniscal lesions, and meniscal lesions are precursors to OA progression.
ICT in promoting the health of the elderly
Kankaala Erik, Oulun yliopisto
The first goal of this thesis was to perform a literary survey on how widely mobile applications are used in detecting signs of depression and memory problems. The second goal was to determine the current widely used tests in detecting mood and cognitive abilities. The third goal was to consider the meaning of equality of knowledge amongst those who have functional disabilities in the information society. The fourth goal was to deploy a pilot questionnaire to the elderly by applying the AWARE Framework environment.
The literary survey was performed by conducting a search in Pubmed and going through and choosing the articles to include. The included articles were analyzed on whether it presented the use of a mobile application or a traditional one. The current widely used tests in detecting mood and cognitive abilities were found by discussions with professionals in the field of healthcare research and also by doing research in the literature and other information found on the internet. The pilot questionnaire was constructed by using the AWARE Framework platform and its ESM-message building feature. Four mobile phones were acquired from the university for two weeks to enable the pilot questionnaire. The themes in the questionnaire were decided upon research over the areas of everyday living that are affected by depression and cognitive decline causing memory problems.
The literary survey concluded that mobile applications are used in detecting signs of depression and memory problems but less widely as are traditional face-to-face and via telephone applications. The widely used tests were chosen to be MMSE, BDI and CERAD. The meaning of equality of knowledge was observed from different angles and concluded in raising a doubt over possible inequality amongst those who have functional disabilities in the information society. The pilot of the questionnaire to the elderly was successful without any unexpected problems. Information was gathered during the pilot including the poll to perform assessment of usability. Based on the assessment the questionnaire-format was applicable on both the question-parts and the usability of the AWARE Framework environment. The results of this work will be applied in the pilot of the project in an actual senior home environment.
Hoitokohteen liikekorjauksen tarkkuus eturauhasen robottisädehoidossa
Korhonen Miitu, Itä-Suomen yliopisto
Eturauhassyöpä on miesten yleisin syöpä. Suomessa todetaan lähes 5000 uutta syöpätapausta vuosittain ja väestön ikääntyessä uusien syöpätapausten määrä tulee kasvamaan entisestään. Eturauhassyövän hoito on hyvin potilasriippuvainen ja vaihtelee seurannasta leikkaus- ja sädehoitoon. Radiokirurgian käyttöön suunniteltu CyberKnife-sädehoitolaite (CK) mahdollistaa hoitokohteen liikkeen lähes reaaliaikaisen seurannan sekä havaitun liikkeen automaattisen kompensoinnin perustuen eturauhaseen implantoituihin markkereihin. Markkereiden lukumäärän ja sijainnin vaikutuksista hoidon tarkkuuteen eturauhassyövän sädehoidossa ei kuitenkaan ole aiemmin raportoitu. Pro gradu -tutkielman päätavoitteena oli määrittää jyväkonfiguraatioiden vaikutukset CK:n liikekorjauksen tarkkuuteen sekä selvittää hoitokohteen liikkeiden automaattisen korjauksen vaikutus hoitokohteen saamaan sädeannokseen.
Mittaukset suoritettiin Kuopion yliopistollisessa sairaalassa (Sädehoitoyksikkö) käyttäen PMMA:sta (polymetyylimetakrylaatti) tehtyä kuution muotoista fantomia sekä erilaisia kultajyväkonfiguraatioita (optimaalinen neljän jyvän tapaus sekä kolme kliinisesti merkittävää tapausta: neljällä, kolmella ja kahdella jyvällä) sen sisällä. Eturauhasen liikkeitä mukailtiin sekä siirtämällä että kallistamalla fantomia kolmeen eri asentoon hoitopöydän avulla (ei liikettä, tyypillinen kliininen liike ja automaattisen liikkeen tunnistuksen hyväksymät maksimaaliset liikkeet). Fantomi säteilytettiin käyttäen tyypillistä eturauhasen hoitosuunnitelmaa (5x7.25 Gy) CK-sädehoitolaitteella. Säteilytetyt annosjakaumat määritettiin koronaalija sagittaalitasoissa käyttäen kuution sisään asetettuja radiokromisia (EBT3) filmejä. Lopuksi filmit skannattiin 72 tuntia säteilytyksen jälkeen ja eri jyväkonfiguraatioilla mitattuja annosjakaumia verrattiin laskennalliseen annosjakaumaan sekä mitattuun optimitapauksen jakaumaan. Annosjakaumien vertailuun käytettiin gamma-analyysiä, käyttäen 5 %/2 mm kriteeriä.
Fraktion aikaisen hoitokohteen liikkeen automaattisen korjauksen tarkkuus oli kliinisesti hyväksyttävällä tasolla, kun kolme tai neljä kultajyvää oli käytössä liikkeen havaitsemiseksi (Taulukko 1). Fantomin liikkeiden ei havaittu merkittävästi muuttavan gamma-laskennan läpäisyosuutta. Kliinisesti hyväksymättömät gamma-laskennan läpäisyosuudet havaittiin vain, kun liikkeen havainnointiin käytettiin kahta kultajyvää.
Pro gradu -tutkielman perusteella voidaan sanoa, että käytettäessä kolmea tai neljää kultajyvää eturauhasen sädehoidossa liikekorjauksen tarkkuus oli kliinisesti riittävä. Optimaalinen neljän jyvän asettelu ei parantanut hoidon tarkkuutta verrattuna tyypillisellä kliinisellä jyväkonfiguraatiolla tai kolmella jyvällä saavutettuun tarkkuuteen. Vain kahden jyvän käyttöä tulisi välttää, sillä niillä ei saavuteta riittävää tarkkuutta hoitokohteen liikkeen korjaamiseksi.
Päivittäisen fyysisen aktiivisuuden suhde kaatumisiin ikääntyneillä
Iida Paksuniemi, Oulun yliopisto
Tämän pro gradu-tutkielman tarkoituksena oli tutkia ikääntyneiden mitatusta päivittäisestä fyysisestä aktiivisuudesta kaatumisia ennustavia tekijöitä. Myös tarkasteluaikaikkunan (epoch) pituuden merkitystä datan analysoinnissa tutkittiin.
Tutkimukseen käytetty kiihtyvyysmittarilla mitattu aktiivisuusdata ja taustatiedot saatiin aiemmin suoritetun SensorBand-kaatumistutkimuksen yhteydessä kerätystä aineistosta. Tähän tutkimukseen otettiin mukaan 14 ikääntynyttä henkilöä, jotka pitivät fyysistä aktiivisuutta mittaavaa kiihtyvyysmittaria. Testihenkilöt olivat iältään 80-92 vuotiaita. Aktiivisuusdatasta analysoitiin päivittäinen keskiarvo ja varianssi vuorokauden eriaikoina (aamuyö, päivä, ilta ja yö) 1 minuutin ja 30 minuutin tarkasteluaikaikkunoissa. Valituista muuttujista kaatumisia ennustavia tekijöitä etsittiin vertaamalla viikkoa ja päivää, jolloin kaatuminen oli tapahtunut, sitä edeltävään ja seuraavaan viikkoon ja päivään. Lisäksi kaatujien ja ei-kaatujien fyysisiä aktiivisuuksia verrattiin toisiinsa.
Ei-kaatujien ja kaatujien viikoittaisista päiväajan fyysisistä aktiivisuuksista löytyi merkitsevää tilastollista eroa. Muissa vuorokauden ajoissa eroja ei löytynyt. Kaatujien fyysisistä aktiivisuuksista viikko- tai päivätasolla ei löytynyt kuitenkaan kaatumisia ennustavia tekijöitä.
Kaatujien perusliikkuminen ei muuttunut kaatumisista johtuen lukuun ottamatta yhtä testihenkilöä, jonka fyysinen aktiivisuus laski mittausjakson ensimmäisen kaatumisen jälkeen. Kaatujat liikkuivat keskimäärin vähemmän kuin ei-kaatujat.
Noise optimization of multi-layer insulation in liquid-helium cryostat for brain imaging
Tervo Aino, Aalto-yliopisto
Introduction: Superconducting quantum interference devices (SQUIDs) can be used as extremely sensitive magnetic-field detectors. They are widely used in biomedical applications such as ultra-low-field magnetic resonance imaging and magnetoencephalography, providing anatomical or functional information of the human brain. To be operated in a superconducting state, the SQUIDs are immersed into a liquid-helium bath, which is housed by a dewar. In order to keep the helium liquefied, the dewar needs thermal insulation against heat transfer by thermal radiation. The insulation is usually implemented by placing aluminized layers into the vacuum space of the dewar. However, the aluminized insulation layers cause thermal magnetic noise due to the thermal motion of free electrons within the conductors. This noise can be a limiting factor if not carefully considered. Based on the literature, thermal magnetic noise can be reduced by breaking the metallic layers into small electrically isolated areas. However, there are currently no detailed studies about the effect of the size of the isolated area or the number of the insulation layers on the noise level. To respond to this lack of information, this study describes methods for accurate calculation of the noise arising from patched superinsulation. In addition, the corresponding boil-off rates are estimated.
Methods: In the simulations, the multi-layer insulation is modeled as patched aluminum sheets, i.e., thin aluminum layers are cut into rectangular areas with a tiny gap between the patches acting as an electrical insulation. The thermal noise currents on the patches are expressed as a linear combination of surface current-density patterns, from which the thermal magnetic noise can be calculated. Boil-off rates are estimated using heat transport equations for radiative heat transfer. The simulations are carried out with numerical tools in Python 3 based on the NumPy and SciPy packages.
Results: The simulations revealed the effects of the isolated area and the number of layers on the noise and the boil-off. The results showed that the patches should be made very small in order to reach sufficiently low noise level. However, small patch size leads to a relatively high helium boiloff rate. The results for magnetic noise were in line with the theoretical values and approximations presented in the literature.
Conclusion: In this study, the thermal magnetic noise arising from the multi-layer insulation and the helium boil-off were simulated in order to ease the design of a new measurement dewar. The presented model was validated to be suitable for analyzing the thermal magnetic noise. According to the results, to reach the desired noise level, the isolated areas should be made very small. Since this might not be practical to manufacture, also other insulation materials should be considered.
Monitoring of postoperative myocardial ischemia and infarction with continuous ECG measurements
Vanhatalo Joonas, Tampereen teknillinen yliopisto
Postoperative myocardial ischemia leading to myocardial infarction(PMI) after a non-cardiac surgery is shown to be common in selected patient population and is associated with increased mortality and morbidity in-hospital and long term [2, 6, 3]. The detection of PMI is challenging as the common symptoms, such as chest pain, are not felt by the patient due to postoperative pain medication. This complicates making the conventional clinical diagnosis of myocardial infarction as according to current criteria for diagnosis a second piece of evidence is needed [5]. Primary and a relatively sensitive detection rate can be achieved using assays of a biomarker called cardiac troponin in order to indicate myocardial injury [4]. However, for troponin concentration to rise an injury must have already occured. Electrocardiography(ECG) changes on the other hand are known to be present in the setting of myocardial ischemia immediately after diminished or restricted blood flow. In addition major part of the postoperative ischemic events are asymptomatic and are thus further stressing the importance of ECG monitoring [1]. Despite the abundant clinical evidence supporting these facts the standard of practice in many hospitals does not include routine troponin assays or even routine ECG monitoring after non-cardiac surgery. Even though many hospitals practice taking sample ECG’s in these situations, still the dynamic nature of the ischemic events might cause this static snapshot to miss important information and the underlying, possible deteriorating, ECG trend. Thus in order to increase the likelihood of timely detection of ischemic events, continuous ECG monitoring in parallel with regular troponin assays should be performed to prevent the ischemia developing further in the first place. At the same time this should be done efficiently with minimal burden for the patient. This study aims to support these statements. Recent advances in several technology areas are driving patient monitoring towards wireless approach to which this scenario of monitoring also fits well.
In the study the ischemic events of patients undergoing a non-cardiac vascular surgery were studied in the Helsinki university central hospital(HUCH). ECG was measured using a Holter ECG system in order to imitate a wireless monitoring setting. The monitoring time was 72 hours after a non-cardiac vascular surgery. Additionally standard 12-lead snapshot ECG’s were measured for comparison to determine the benefits of continuous ECG monitoring over the snapshot ECG’s. The performance of the continuous measurements was used to assess the quality of the measurements for future development of ischemia monitoring. Statistical and other data analysis was mostly done using MATLAB 2015b. Furthermore a custom software analysis tool was developed in order to visualize and utilize the data in an orderly manner. The tool was created using C#(C-sharp)- programming language. The tool was used to make visual annotations and to post-process the data as well as a platform to be used in further development of ischemia monitoring.
From the study group of 56 patients, 19 were diagnosed to have ischemia. Out of the 19 patients, 5 were diagnosed to have myocardial infarction during the postoperative monitoring. All ischemia appeared as depression of the ST-segment. After analysing the data the ECG lead sensitivities were found to be consistent with similar studies and clinical guidelines. Moreover the ischemic burden revealed by continuous monitoring was considerable and increased cumulatively during the monitoring. Additionally the first indications of ischemia were seen at the latest or before the snapshot ECG’s.
The thesis shows evidence that continuous realtime ECG monitoring of certain postoperative patients would reduce the total ischemic burden of patients undergone a non-cardiac surgery. The dynamic nature of postoperative myocardial ischemia demand continuous monitoring, and a low sensitivity is achieved when only snapshot ECG’s are used. Because of the restrictions set by wired measurements in such a scenario, continuous monitoring should be done using wireless measurements.
Static and dynamic approaches for pain detection and pain intensity estimation using facial expression
Zebarjadi Niloufar, Tampereen teknillinen yliopisto
Self-reported pain is the most convensional way of assessing pain intensity in hospitals, while in many circumstances is not an accurate measure or not even possible to measure (e.g. in intensive care units). Continuous and automatic pain level evaluation is an advantageous method to overcome this issue. Our aim in this study is to propose a new technique using facial expression to automatically determine pain intensity levels and to provide comprehensive evaluation within two dimensional (2D) and three dimensional (3D) feature descriptors for solving this issue.
In the experiment, we conduct our technique on UNBC-McMaster shoulder pain expression archive database containing 200 video sequences of spontaneous facial expressions (48398 frames) of 25 patients suffering from shoulder pain. The framework of study is: Preprocessing of facial to remove the distracting information such as background and extracting the region of interest (face) in three stages of face tracking, face registration and face cropping. In the next step, we employ two frequently used static feature descriptors for analysis of facial expression (local binary pattern (LBP) and dense scale-invariant feature transform (DSIFT)) as well as their dynamic features (local binary patterns on three orthogonal planes (LBP-TOP) and three dimensional Scale-Invariant Feature Transform (3D-SIFT)) to have comprehensive evaluation within two dimensional (2D) and three dimensional (3D) feature descriptors. Afterwards, support vector regression (SVR) is applied in this study to predict pain intensity level.
Optimal block size is required to be selected for all feature descriptors. Wide range of parameters are experimented through an iterative search algorithm to find efficient values for feature block size and SVR parameters. The table represents the best MSE and PCC results of all static and dynamic descriptors on the complete UNBC-McMaster shoulder pain expression archive database. Best parameters of the features are provided as lower indices in this table. Parameters for both LBP and LBP-TOP are number of neighboring points around each central pixel(P), radius around each central pixel(R), number of division in row (NDR), number of division in column(NDC), number of division in time(NDT),respectively. In 2D and 3D-SIFT, parameters are: size of the extracted descriptor(S) number of bin in the X direction(NX), number of bin in the Y direction(NY) and number of division in time(NT).
In conclusion, from the least MSE point of view, dynamic features including LBP-TOP and 3D-SIFT surpass the static feature descriptors including LBP and DSIFT for pain intensity estimation. Nevertheless, considering acquired PCC values, LBP family are leading to superior results compared to the SIFT family. Interestingly, with respect to both of the metrics, temporal feature descriptors in either of the feature families outperforms the static feature descriptors of the same family. The reason is that, there are useful temporal information present in the sequences which are exceeding the performance of regression machine and those information might lost by employing static feature descriptors. Although our obtained results are limited to UNBC-McMaster shoulder pain expression archive database, they are in agreement with in this context. Comparing 3D descriptors performances attained in our experiments, by either of the metrics, LBP-TOP gives better results than 3D-SIFT. The same statement can be proposed for the corresponding 2D descriptors, which LBP wins DSIFT model. This outcome shows the advantage of LBP family on the facial expression applications and is correlated with the results obtained in many papers contributed in facial expression applications.