Browsing by Subject "SEGMENTATION"

Sort by: Order: Results:

Now showing items 1-20 of 25
  • Kaipainen, Aku L.; Pitkänen, Johanna; Haapalinna, Fanni; Jääskeläinen, Olli; Jokinen, Hanna; Melkas, Susanna; Erkinjuntti, Timo; Vanninen, Ritva; Koivisto, Anne M.; Lötjönen, Jyrki; Koikkalainen, Juha; Herukka, Sanna-Kaisa; Julkunen, Valtteri (2021)
    Purpose Automated analysis of neuroimaging data is commonly based on magnetic resonance imaging (MRI), but sometimes the availability is limited or a patient might have contradictions to MRI. Therefore, automated analyses of computed tomography (CT) images would be beneficial. Methods We developed an automated method to evaluate medial temporal lobe atrophy (MTA), global cortical atrophy (GCA), and the severity of white matter lesions (WMLs) from a CT scan and compared the results to those obtained from MRI in a cohort of 214 subjects gathered from Kuopio and Helsinki University Hospital registers from 2005 - 2016. Results The correlation coefficients of computational measures between CT and MRI were 0.9 (MTA), 0.82 (GCA), and 0.86 (Fazekas). CT-based measures were identical to MRI-based measures in 60% (MTA), 62% (GCA) and 60% (Fazekas) of cases when the measures were rounded to the nearest full grade variable. However, the difference in measures was 1 or less in 97-98% of cases. Similar results were obtained for cortical atrophy ratings, especially in the frontal and temporal lobes, when assessing the brain lobes separately. Bland-Altman plots and weighted kappa values demonstrated high agreement regarding measures based on CT and MRI. Conclusions MTA, GCA, and Fazekas grades can also be assessed reliably from a CT scan with our method. Even though the measures obtained with the different imaging modalities were not identical in a relatively extensive cohort, the differences were minor. This expands the possibility of using this automated analysis method when MRI is inaccessible or contraindicated.
  • Acosta, H.; Tuulari, J. J.; Kantojärvi, K.; Lewis, J. D.; Hashempour, N.; Scheinin, N. M.; Lehtola, S. J.; Fonov, V. S.; Collins, D. L.; Evans, A.; Parkkola, R.; Lahdesmaki, T.; Saunavaara, J.; Merisaari, H.; Karlsson, L.; Paunio, T.; Karlsson, H. (2021)
    Genetic variants in the oxytocin receptor (OTR) have been linked to distinct social phenotypes, psychiatric disorders and brain volume alterations in adults. However, to date, it is unknown how OTR genotype shapes prenatal brain development and whether it interacts with maternal prenatal environmental risk factors on infant brain volumes. In 105 Finnish mother-infant dyads (44 female, 11-54 days old), the association of offspring OTR genotype rs53576 and its interaction with prenatal maternal anxiety (revised Symptom Checklist 90, gestational weeks 14, 24, 34) on infant bilateral amygdalar, hippocampal and caudate volumes were probed. A sex-specific main effect of rs53576 on infant left hippocampal volumes was observed. In boys compared to girls, left hippocampal volumes were significantly larger in GG-homozygotes compared to A-allele carriers. Furthermore, genotype rs53576 and prenatal maternal anxiety significantly interacted on right hippocampal volumes irrespective of sex. Higher maternal anxiety was associated both with larger hippocampal volumes in A allele carriers than GG-homozygotes, and, though statistically weak, also with smaller right caudate volumes in GG-homozygotes than A-allele carriers. Our study results suggest that OTR genotype enhances hippocampal neurogenesis in male GG-homozygotes. Further, prenatal maternal anxiety might induce brain alterations that render GG-homozygotes compared to A-allele carriers more vulnerable to depression.
  • Burunat, Iballa; Brattico, Elvira; Puoliväli, Tuomas; Ristaniemi, Tapani; Sams, Mikko; Toiviainen, Petri (2015)
    Musical training leads to sensory and motor neuroplastic changes in the human brain. Motivated by findings on enlarged corpus callosum in musicians and asymmetric somatomotor representation in string players, we investigated the relationship between musical training, callosal anatomy, and interhemispheric functional symmetry during music listening. Functional symmetry was increased in musicians compared to nonmusicians, and in keyboardists compared to string players. This increased functional symmetry was prominent in visual and motor brain networks. Callosal size did not significantly differ between groups except for the posterior callosum in musicians compared to nonmusicians. We conclude that the distinctive postural and kinematic symmetry in instrument playing cross-modally shapes information processing in sensory-motor cortical areas during music listening. This cross-modal plasticity suggests that motor training affects music perception.
  • Hollandi, Reka; Diosdi, Akos; Hollandi, Gabor; Moshkov, Nikita; Horvath, Peter (2020)
    AnnotatorJ combines single-cell identification with deep learning (DL) and manual annotation. Cellular analysis quality depends on accurate and reliable detection and segmentation of cells so that the subsequent steps of analyses, for example, expression measurements, may be carried out precisely and without bias. DL has recently become a popular way of segmenting cells, performing unimaginably better than conventional methods. However, such DL applications may be trained on a large amount of annotated data to be able to match the highest expectations. High-quality annotations are unfortunately expensive as they require field experts to create them, and often cannot be shared outside the lab due to medical regulations. We propose AnnotatorJ, an ImageJ plugin for the semiautomatic annotation of cells (or generally, objects of interest) on (not only) microscopy images in 2D that helps find the true contour of individual objects by applying U-Net-based presegmentation. The manual labor of hand annotating cells can be significantly accelerated by using our tool. Thus, it enables users to create such datasets that could potentially increase the accuracy of state-of-the-art solutions, DL or otherwise, when used as training data.
  • Ng, Wai Tong; But, Barton; Choi, Horace C. W.; de Bree, Remco; Lee, Anne W. M.; Lee, Victor H. F.; Lopez, Fernando; Mäkitie, Antti A.; Rodrigo, Juan P.; Saba, Nabil F.; Tsang, Raymond K. Y.; Ferlito, Alfio (2022)
    Introduction: Nasopharyngeal carcinoma (NPC) is endemic to Eastern and South-Eastern Asia, and, in 2020, 77% of global cases were diagnosed in these regions. Apart from its distinct epidemiology, the natural behavior, treatment, and prognosis are different from other head and neck cancers. With the growing trend of artificial intelligence (AI), especially deep learning (DL), in head and neck cancer care, we sought to explore the unique clinical application and implementation direction of AI in the management of NPC. Methods: The search protocol was performed to collect publications using AI, machine learning (ML) and DL in NPC management from PubMed, Scopus and Embase. The articles were filtered using inclusion and exclusion criteria, and the quality of the papers was assessed. Data were extracted from the finalized articles. Results: A total of 78 articles were reviewed after removing duplicates and papers that did not meet the inclusion and exclusion criteria. After quality assessment, 60 papers were included in the current study. There were four main types of applications, which were auto-contouring, diagnosis, prognosis, and miscellaneous applications (especially on radiotherapy planning). The different forms of convolutional neural networks (CNNs) accounted for the majority of DL algorithms used, while the artificial neural network (ANN) was the most frequent ML model implemented. Conclusion: There is an overall positive impact identified from AI implementation in the management of NPC. With improving AI algorithms, we envisage AI will be available as a routine application in a clinical setting soon.
  • Guirado, Ramon; Carceller, Hector; Castillo-Gomez, Esther; Castren, Eero; Nacher, Juan (2018)
    The quantification of the expression of different molecules is a key question in both basic and applied sciences. While protein quantification through molecular techniques leads to the loss of spatial information and resolution, immunohistochemistry is usually associated with time-consuming image analysis and human bias. In addition, the scarce automatic software analysis is often proprietary and expensive and relies on a fixed threshold binarization. Here we describe and share a set of macros ready for automated fluorescence analysis of large batches of fixed tissue samples using FIJI/ImageJ. The quantification of the molecules of interest are based on an automatic threshold analysis of immunofluorescence images to automatically identify the top brightest structures of each image. These macros measure several parameters commonly quantified in basic neuroscience research, such as neuropil density and fluorescence intensity of synaptic puncta, perisomatic innervation and col-localization of different molecules and analysis of the neurochemical phenotype of neuronal subpopulations. In addition, these same macro functions can be easily modified to improve similar analysis of fluorescent probes in human biopsies for diagnostic purposes based on the expression patterns of several molecules.
  • Kwak, Gloria Hyunjung; Kwak, Eun-Jung; Song, Jae Min; Park, Hae Ryoun; Jung, Yun-Hoa; Cho, Bong-Hae; Hui, Pan; Hwang, Jae Joon (2020)
    The practicability of deep learning techniques has been demonstrated by their successful implementation in varied fields, including diagnostic imaging for clinicians. In accordance with the increasing demands in the healthcare industry, techniques for automatic prediction and detection are being widely researched. Particularly in dentistry, for various reasons, automated mandibular canal detection has become highly desirable. The positioning of the inferior alveolar nerve (IAN), which is one of the major structures in the mandible, is crucial to prevent nerve injury during surgical procedures. However, automatic segmentation using Cone beam computed tomography (CBCT) poses certain difficulties, such as the complex appearance of the human skull, limited number of datasets, unclear edges, and noisy images. Using work-in-progress automation software, experiments were conducted with models based on 2D SegNet, 2D and 3D U-Nets as preliminary research for a dental segmentation automation tool. The 2D U-Net with adjacent images demonstrates higher global accuracy of 0.82 than naïve U-Net variants. The 2D SegNet showed the second highest global accuracy of 0.96, and the 3D U-Net showed the best global accuracy of 0.99. The automated canal detection system through deep learning will contribute significantly to efficient treatment planning and to reducing patients’ discomfort by a dentist. This study will be a preliminary report and an opportunity to explore the application of deep learning to other dental fields.
  • Kuula, Juho; Martola, Juha; Hakkarainen, Antti; Räikkönen, Katri; Savolainen, Sauli; Salli, Eero; Hovi, Petteri; Björkqvist, Johan; Kajantie, Eero; Lundbom, Nina (2022)
    Objectives To assess radiographic brain abnormalities and investigate volumetric differences in adults born preterm at very low birth weight ( Study design We recruited 79 adult same-sex sibling pairs with one born preterm at very low birth weight and the sibling at term. We acquired 3-T brain magnetic resonance imaging from 78 preterm participants and 72 siblings. A neuroradiologist, masked to participants' prematurity status, reviewed the images for parenchymal and structural abnormalities, and FreeSurfer software 6.0 was used to conduct volumetric analyses. Data were analyzed by linear mixed models. Results We found more structural abnormalities in very low birth weight participants than in siblings (37% vs 13%). The most common finding was periventricular leukomalacia, present in 15% of very low birth weight participants and in 3% of siblings. The very low birth weight group had smaller absolute brain volumes (-0.4 SD) and, after adjusting for estimated intracranial volume, less gray matter (-0.2 SD), larger ventricles (1.5 SD), smaller thalami (-0.6 SD), caudate nuclei (-0.4 SD), right hippocampus (-0.4 SD), and left pallidum (-0.3 SD). We saw no volume differences in total white matter (-0.04 SD; 95% CI, -0.13 to 0.09). Conclusions Preterm very low birth weight adults had a higher prevalence of brain abnormalities than their termborn siblings. They also had smaller absolute brain volumes, less gray but not white matter, and smaller volumes in several gray matter structures.
  • Claesson, Tor-björn; Putaala, Jukka; Shams, Sara; Salli, Eero; Gordin, Daniel; Liebkind, Ron; Forsblom, Carol; Summanen, Paula A.; Tatlisumak, Turgut; Groop, Per-Henrik; Martola, Juha; Thorn, Lena M. (2020)
    Background and purpose: Degenerative change of the corpus callosum might serve as a clinically useful surrogate marker for net pathological cerebral impact of diabetes type 1. We compared manual and automatic measurements of the corpus callosum, as well as differences in callosal cross-sectional area between subjects with type 1 diabetes and healthy controls. Materials and methods: This is a cross-sectional study on 188 neurologically asymptomatic participants with type 1 diabetes and 30 healthy age- and sex-matched control subjects, recruited as part of the Finnish Diabetic Nephropathy Study. All participants underwent clinical work-up and brain MRI. Callosal area was manually measured and callosal volume quantified with FreeSurfer. The measures were normalized using manually measured mid-sagittal intracranial area and volumetric intracranial volume, respectively. Results: Manual and automatic measurements correlated well (callosal area vs. volume: rho = 0.83, p <0.001 and mid-sagittal area vs. intracranial volume: rho = 0.82, p <0.001). We found no significant differences in the callosal measures between cases and controls. In type 1 diabetes, the lowest quartile of normalized callosal area was associated with higher insulin doses (p = 0.029) and reduced insulin sensitivity (p = 0.033). In addition, participants with more than two cerebral microbleeds had smaller callosal area (p = 0.002). Conclusion: Manually measured callosal area and automatically segmented are interchangeable. The association seen between callosal size with cerebral microbleeds and insulin resistance is indicative of small vessel disease pathology in diabetes type 1.
  • Hamedianfar, Alireza; Mohamedou, Cheikh; Kangas, Annika; Vauhkonen, Jari (2022)
    Data processing for forestry applications is challenged by the increasing availability of multi-source and multi-temporal data. The advancements of Deep Learning (DL) algorithms have made it a prominent family of methods for machine learning and artificial intelligence. This review determines the current state-of-the-art in using DL for solving forestry problems. Although DL has shown potential for various estimation tasks, the applications of DL to forestry are in their infancy. The main study line has related to comparing various Convolutional Neural Network (CNN) architectures between each other and against more shallow machine learning techniques. The main asset of DL is the possibility to internally learn multi-scale features without an explicit feature extraction step, which many people typically perceive as a black box approach. According to a comprehensive literature review, we identified challenges related to (1) acquiring sufficient amounts of representative and labelled training data, (2) difficulties to select suitable DL architecture and hyperparameterization among many methodological choices and (3) susceptibility to overlearn the training data and consequent risks related to the generalizability of the predictions, which can however be reduced by proper choices on the above. We recognized possibilities in building time-series prediction strategies upon Recurrent Neural Network architectures and, more generally, re-thinking forestry applications in terms of components inherent to DL. Nevertheless, DL applications remain data-driven, in contrast to being based on causal reasoning, and currently lack many best practices of conventional forestry modelling approaches. The benefits of DL depend on the application, and the practitioners are advised to ex ante subject their requirements to operational data availability, for example. By this review, we contribute to the technical discussion about the prospects of DL for forestry and shed light on properties that require attention from the practitioners.
  • Tanhuanpää, Topi; Yu, Xiaowei; Luoma, Ville; Saarinen, Ninni; Raisio, Juha; Hyyppä, Juha; Kumpula, Timo; Holopainen, Markus (2019)
    Urban forests consist of patches of recreational areas, parks, and single trees on roadsides and other forested urban areas. Large number of tree species and heterogeneous growing conditions result in diverse canopy structure. High variation can be found both at level of single tree crowns and in canopy characteristics of larger areas. As urban forests are typically managed with small-scale, even tree-level operations, there is a need for detailed forest information. In this study, the effect of varying canopy conditions was tested on nine individual tree detection (ITD) methods. All methods utilized airborne laser scanning (ALS)-derived canopy height models (CHM) and different modifications of watershed segmentation (WS). The performance of mapping methods was compared in three strata with varying mean height and canopy cover. The results showed considerable variation between the methods when tested in varying canopy conditions. Especially, presence of large broadleaved trees affected the accuracy of detecting individual trees. The best performing methods for the three strata were G0.7, F2 and Gadapt. The areas with low canopy cover turned out problematic for all ITD methods tested as co-occurrence of small trees and large deciduous trees affected the accuracy significantly. Overall, The results show that stratification can be used to enhance the quality of ITD in urban park areas. However, heterogeneous canopy structure and varying growth patterns typical for urban parks lower the accuracy of tree detection. Also, according to our results we suggest that canopy height and canopy cover alone are insufficient attributes for stratifying urban canopy conditions.
  • Zhou, Guangyu; Hotta, Jaakko; Lehtinen, Maria K.; Forss, Nina; Hari, Riitta (2015)
    The choroid plexus, located in brain ventricles, has received surprisingly little attention in clinical neuroscience. In morphometric brain analysis, we serendipitously found a 21% increase in choroid plexus volume in 12 patients suffering from complex regional pain syndrome (CRPS) compared with age- and gender-matched healthy subjects. No enlargement was observed in a group of 8 patients suffering from chronic pain of other etiologies. Our findings suggest involvement of the choroid plexus in the pathogenesis of CRPS. Since the choroid plexus can mediate interaction between peripheral and brain inflammation, our findings pinpoint the choroid plexus as an important target for future research of central pain mechanisms.
  • Kärhä, Kalle; Nurmela, Sari; Karvonen, Heikki; Kivinen, Veli-Pekka; Melkas, Timo; Nieminen, Miikka (2019)
    Trestima Stack is a mobile application innovated by Trestima Ltd. It is based on machine vision, which measures the volume of a timber stack from images taken by a smartphone or a tablet device. The aim of this study was to determine the accuracy (i.e. measurement difference) and effective measurement time consumption of the Trestima Stack application compared to a conventional stacked timber measurement method. Research data consisted of a total of 60 timber stacks, of which 32 were measured in terminal and intermediate yards and 28 at roadside landings. The control volumes of the stacks were measured in September 2016 - January 2017 at the Stora Enso Anjala, Imatra and Varkaus mills by hydrostatic weighting. The total control volume of pulpwood in the study was 11,957 m(3) solid over the bark (m(3)). Across all study data, the accuracy of Trestima Stack averaged + 2.7%. In large terminal yards, accuracy was better (+ 0.7%) than at smaller roadside landings (+ 4.5%), whereas with the conventional stacked timber measurement method, the measurement accuracy was at a similar level in terminal yards (- 4.8%) as at roadside landings (- 4.9%). There was a statistically significant difference between the measurement methods used in measurement accuracy. The most common reason for inaccuracy with the Trestima Stack application was empty space in the final image framing around the stack. The average effective total measurement time consumption with Trestima Stack was 10.6 s/m(3), while it was 13.7 s/m(3) with the conventional stacked timber measurement method. For both measurement methods, there was a statistically significant negative correlation between the stack size and the volume-based effective total measurement time consumption. On the basis of this study, the Trestima Stack application can be recommended for inventorying timber stacks at the roadside landings, particularly when the stacks measured consist of several measurement batches.
  • Lember, Juri; Gasbarra, Dario; Koloydenko, Alexey; Kuljus, Kristi (2019)
    The article studies different methods for estimating the Viterbi path in the Bayesian framework. The Viterbi path is an estimate of the underlying state path in hidden Markov models (HMMs), which has a maximum joint posterior probability. Hence it is also called the maximum a posteriori (MAP) path. For an HMM with given parameters, the Viterbi path can be easily found with the Viterbi algorithm. In the Bayesian framework the Viterbi algorithm is not applicable and several iterative methods can be used instead. We introduce a new EM-type algorithm for finding the MAP path and compare it with various other methods for finding the MAP path, including the variational Bayes approach and MCMC methods. Examples with simulated data are used to compare the performance of the methods. The main focus is on non-stochastic iterative methods and our results show that the best of those methods work as well or better than the best MCMC methods. Our results demonstrate that when the primary goal is segmentation, then it is more reasonable to perform segmentation directly by considering the transition and emission parameters as nuisance parameters.
  • Anurova, Irina; Vetchinnikova, Svetlana; Dobrego, Aleksandra; Williams, Nitin; Mikusova, Nina; Suni, Antti; Mauranen, Anna; Palva, Satu (2022)
    Chunking language has been proposed to be vital for comprehension enabling the extraction of meaning from a continuous stream of speech. However, neurocognitive mechanisms of chunking are poorly understood. The present study investigated neural correlates of chunk boundaries intuitively identified by listeners in natural speech drawn from linguistic corpora using magneto-and electroencephalography (MEEG). In a behavioral experiment, subjects marked chunk boundaries in the excerpts intuitively, which revealed highly consistent chunk boundary markings across the subjects. We next recorded brain activity to investigate whether chunk boundaries with high and medium agreement rates elicit distinct evoked responses compared to non-boundaries. Pauses placed at chunk boundaries elicited a closure positive shift with the sources over bilateral auditory cortices. In contrast, pauses placed within a chunk were perceived as interruptions and elicited a biphasic emitted potential with sources located in the bilateral primary and non-primary auditory areas with right-hemispheric dominance, and in the right inferior frontal cortex. Furthermore, pauses placed at stronger boundaries elicited earlier and more prominent activation over the left hemisphere suggesting that brain responses to chunk boundaries of natural speech can be modulated by the relative strength of different linguistic cues, such as syntactic structure and prosody.
  • Wang, Sheng H.; Lobier, Muriel; Siebenhühner, Felix; Puoliväli, Tuomas; Palva, Satu; Palva, J. Matias (2018)
    It has not been well documented that MEG/EEG functional connectivity graphs estimated with zero-lag-free interaction metrics are severely confounded by a multitude of spurious interactions (SI), i.e., the false-positive “ghosts” of true interactions [1,2]. These SI are caused by the multivariate linear mixing between sources, and thus they pose a severe challenge to the validity of connectivity analysis. Due to the complex nature of signal mixing and the SI problem, there is a need to intuitively demonstrate how the SI are discovered and how they can be attenuated using a novel approach that we termed hyperedge bundling. Here we provide a dataset with software with which the readers can perform simulations in order to better understand the theory and the solution to SI. We include the supplementary material of [1] that is not directly relevant to the hyperedge bundling per se but reflects important properties of the MEG source model and the functional connectivity graphs. For example, the gyri of dorsal-lateral cortices are the most accurately modeled areas; the sulci of inferior temporal, frontal and the insula have the least modeling accuracy. Importantly, we found the interaction estimates are heavily biased by the modeling accuracy between regions, which means the estimates cannot be straightforwardly interpreted as the coupling between brain regions. This raise a red flag that the conventional method of thresholding graphs by estimate values is rather suboptimal: because the measured topology of the graph reflects the geometric property of source-model instead of the cortical interactions under investigation.
  • Bosseler, Alexis N.; Teinonen, Tuomas; Tervaniemi, Mari; Huotilainen, Minna (2016)
    Statistical learning and the social contexts of language addressed to infants are hypothesized to play important roles in early language development. Previous behavioral work has found that the exaggerated prosodic contours of infant-directed speech (IDS) facilitate statistical learning in 8-month-old infants. Here we examined the neural processes involved in on-line statistical learning and investigated whether the use of IDS facilitates statistical learning in sleeping newborns. Event-related potentials (ERPs) were recorded while newborns were exposed to 12 pseudo-words, six spoken with exaggerated pitch contours of IDS and six spoken without exaggerated pitch contours (ADS) in ten alternating blocks. We examined whether ERP amplitudes for syllable position within a pseudo-word (word-initial vs. word-medial vs. word-final, indicating statistical word learning) and speech register (ADS vs. IDS) would interact. The ADS and IDS registers elicited similar ERP patterns for syllable position in an early 0-100 ms component but elicited different ERP effects in both the polarity and topographical distribution at 200-400 ms and 450-650 ms. These results provide the first evidence that the exaggerated pitch contours of IDS result in differences in brain activity linked to on-line statistical learning in sleeping newborns.
  • Vuoksimaa, Eero; McEvoy, Linda K.; Holland, Dominic; Franz, Carol E.; Kremen, William S. (2020)
    Mild cognitive impairment (MCI) is a heterogeneous condition with variable outcomes. Improving diagnosis to increase the likelihood that MCI reliably reflects prodromal Alzheimer's Disease (AD) would be of great benefit for clinical practice and intervention trials. In 230 cognitively normal (CN) and 394 MCI individuals from the Alzheimer's Disease Neuroimaging Initiative, we studied whether an MCI diagnostic requirement of impairment on at least two episodic memory tests improves 3-year prediction of medial temporal lobe atrophy and progression to AD. Based on external age-adjusted norms for delayed free recall on the Rey Auditory Verbal Learning Test (AVLT), MCI participants were further classified as having normal (AVLT+, above -1 SD, n = 121) or impaired (AVLT -, -1 SD or below, n = 273) AVLT performance. CN, AVLT+, and AVLT- groups differed significantly on baseline brain (hippocampus, entorhinal cortex) and cerebrospinal fluid (amyloid, tau, p-tau) biomarkers, with the AVLT- group being most abnormal. The AVLT- group had significantly more medial temporal atrophy and a substantially higher AD progression rate than the AVLT+ group (51% vs. 16%, p <0.001). The AVLT+ group had similar medial temporal trajectories compared to CN individuals. Results were similar even when restricted to individuals with above average (based on the CN group mean) baseline medial temporal volume/thickness. Requiring impairment on at least two memory tests for MCI diagnosis can markedly improve prediction of medial temporal atrophy and conversion to AD, even in the absence of baseline medial temporal atrophy. This modification constitutes a practical and cost-effective approach for clinical and research settings.
  • Reijonen, Vappu; Kanninen, Liisa K.; Hippeläinen, Eero; Lou, Yan-Ru; Salli, Eero; Sofiev, Alexey; Malinen, Melina; Paasonen, Timo; Yliperttula, Marjo; Kuronen, Antti; Savolainen, Sauli (2017)
    Purpose: Absorbed radiation dose-response relationships are not clear in molecular radiotherapy (MRT). Here, we propose a voxel-based dose calculation system for multicellular dosimetry in MRT. We applied confocal microscope images of a spherical cell aggregate i.e. a spheroid, to examine the computation of dose distribution within a tissue from the distribution of radiopharmaceuticals. Methods: A confocal microscope Z-stack of a human hepatocellular carcinoma HepG2 spheroid was segmented using a support-vector machine algorithm and a watershed function. Heterogeneity in activity uptake was simulated by selecting a varying amount of the cell nuclei to contain In-111, I-125, or Lu-177. Absorbed dose simulations were carried out using vxlPen, a software application based on the Monte Carlo code PENELOPE. Results: We developed a schema for radiopharmaceutical dosimetry. The schema utilizes a partially supervised segmentation method for cell-level image data together with a novel main program for voxel-based radiation dose simulations. We observed that for 177Lu, radiation cross-fire enabled full dose coverage even if the radiopharmaceutical had accumulated to only 60% of the spheroid cells. This effect was not found with 111In and 125I. Using these Auger/internal conversion electron emitters seemed to guarantee that only the cells with a high enough activity uptake will accumulate a lethal amount of dose, while neighboring cells are spared. Conclusions: We computed absorbed radiation dose distributions in a 3D-cultured cell spheroid with a novel multicellular dosimetric chain. Combined with pharmacological studies in different tissue models, our cell-level dosimetric calculation method can clarify dose-response relationships for radiopharmaceuticals used in MRT. (C) 2017 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.
  • Luotamo, Markku Ilkka Juhana; Metsämäki, Sari; Klami, Arto (2021)
    Semantic segmentation by convolutional neural networks (CNN) has advanced the state of the art in pixel-level classification of remote sensing images. However, processing large images typically requires analyzing the image in small patches, and hence, features that have a large spatial extent still cause challenges in tasks, such as cloud masking. To support a wider scale of spatial features while simultaneously reducing computational requirements for large satellite images, we propose an architecture of two cascaded CNN model components successively processing undersampled and full-resolution images. The first component distinguishes between patches in the inner cloud area from patches at the cloud's boundary region. For the cloud-ambiguous edge patches requiring further segmentation, the framework then delegates computation to a fine-grained model component. We apply the architecture to a cloud detection data set of complete Sentinel-2 multispectral images, approximately annotated for minimal false negatives in a land-use application. On this specific task and data, we achieve a 16% relative improvement in pixel accuracy over a CNN baseline based on patching.