Marker-controlled watershed with deep edge emphasis and optimized H-minima transform for automatic segmentation of densely cultivated 3D cell nuclei.

BACKGROUND: The segmentation of 3D cell nuclei is essential in many tasks, such as targeted molecular radiotherapies (MRT) for metastatic tumours, toxicity screening, and the observation of proliferating cells. In recent years, one popular method for automatic segmentation of nuclei has been deep learning enhanced marker-controlled watershed transform. In this method, convolutional neural networks (CNNs) have been used to create nuclei masks and markers, and the watershed algorithm for the instance segmentation. We studied whether this method could be improved for the segmentation of densely cultivated 3D nuclei via developing multiple system configurations in which we studied the effect of edge emphasizing CNNs, and optimized H-minima transform for mask and marker generation, respectively. RESULTS: The dataset used for training and evaluation consisted of twelve in vitro cultivated densely packed 3D human carcinoma cell spheroids imaged using a confocal microscope. With this dataset, the evaluation was performed using a cross-validation scheme. In addition, four independent datasets were used for evaluation. The datasets were resampled near isotropic for our experiments. The baseline deep learning enhanced marker-controlled watershed obtained an average of 0.69 Panoptic Quality (PQ) and 0.66 Aggregated Jaccard Index (AJI) over the twelve spheroids. Using a system configuration, which was otherwise the same but used 3D-based edge emphasizing CNNs and optimized H-minima transform, the scores increased to 0.76 and 0.77, respectively. When using the independent datasets for evaluation, the best performing system configuration was shown to outperform or equal the baseline and a set of well-known cell segmentation approaches. CONCLUSIONS: The use of edge emphasizing U-Nets and optimized H-minima transform can improve the marker-controlled watershed transform for segmentation of densely cultivated 3D cell nuclei. A novel dataset of twelve spheroids was introduced to the public.

Brain Volumes and Abnormalities in Adults Born Preterm at Very Low Birth Weight.

OBJECTIVES: To assess radiographic brain abnormalities and investigate volumetric differences in adults born preterm at very low birth weight (<1500 g), using siblings as controls. 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 term-born siblings. They also had smaller absolute brain volumes, less gray but not white matter, and smaller volumes in several gray matter structures.

Automatic CT Angiography Lesion Segmentation Compared to CT Perfusion in Ischemic Stroke Detection: a Feasibility Study.

In stroke imaging, CT angiography (CTA) is used for detecting arterial occlusions. These images could also provide information on the extent of ischemia. The study aim was to develop and evaluate a convolutional neural network (CNN)-based algorithm for detecting and segmenting acute ischemic lesions from CTA images of patients with suspected middle cerebral artery stroke. These results were compared to volumes reported by widely used CT perfusion-based RAPID software (IschemaView). A 42-layer-deep CNN was trained on 50 CTA volumes with manually delineated targets. The lower bound for predicted lesion size to reliably discern stroke from false positives was estimated. The severity of false positives and false negatives was reviewed visually to assess the clinical applicability and to further guide the method development. The CNN model corresponded to the manual segmentations with voxel-wise sensitivity 0.54 (95% confidence interval: 0.44-0.63), precision 0.69 (0.60-0.76), and Sørensen-Dice coefficient 0.61 (0.52-0.67). Stroke/nonstroke differentiation accuracy 0.88 (0.81-0.94) was achieved when only considering the predicted lesion size (i.e., regardless of location). By visual estimation, 46% of cases showed some false findings, such as CNN highlighting chronic periventricular white matter changes or beam hardening artifacts, but only in 9% the errors were severe, translating to 0.91 accuracy. The CNN model had a moderately strong correlation to RAPID-reported Tmax > 10 s volumes (Pearson's r = 0.76 (0.58-0.86)). The results suggest that detecting anterior circulation ischemic strokes from CTA using a CNN-based algorithm can be feasible when accompanied with physiological knowledge to rule out false positives.

Rotation and torsion of the left ventricle with cardiovascular magnetic resonance tagging: comparison of two analysis methods.

BACKGROUND: Left ventricle rotation and torsion are fundamental components of myocardial function, and several software packages have been developed for analysis of these components. The purpose of this study was to compare the suitability of two software packages with different technical principles for analysis of rotation and torsion of the left ventricle during systole. METHODS: A group of hypertrophic cardiomyopathy (HCM) patients (N = 14, age 43 ± 11 years), mutation carriers without hypertrophy (N = 10, age 34 ± 13 years), and healthy relatives (N = 12, age 43 ± 17 years) underwent a cardiovascular magnetic resonance examination, including spatial modulation of magnetization tagging sequences in basal and apical planes of the left ventricle. The tagging images were analyzed offline using a harmonic phase image analysis method with Gabor filtering and a non-rigid registration-based free-form deformation technique. Left-ventricle rotation and torsion scores were obtained from end-diastole to end-systole with both software. RESULTS: Analysis was successful in all cases with both software applications. End-systolic torsion values between the study groups were not statistically different with either software. End-systolic apical rotation, end-systolic basal rotation, and end-systolic torsion were consistently higher when analyzed with non-rigid registration than with harmonic phase-based analysis (p <  0.0001). End-systolic rotation and torsion values had significant correlations between the two software (p <  0.0001), most significant in the apical plane. CONCLUSIONS: When comparing absolute values of rotation and torsion between different individuals, software-specific reference values are required. Harmonic phase flow with Gabor filtering and non-rigid registration-based methods can both be used reliably in the analysis of systolic rotation and torsion patterns of the left ventricle.

Potential occupational exposures in diagnostic and interventional radiology: statistical modeling based on Finnish national dose registry data.

BACKGROUND: Radiation worker categorization and exposure monitoring practices must be proportional to the current working environment. PURPOSE: To analyze exposure data of Finnish radiological workers and to estimate the magnitude and frequency of their potential occupational radiation exposure, and to propose appropriate radiation worker categorization. MATERIAL AND METHODS: Estimates of the probabilities of annual effective doses exceeding certain levels were obtained by calculating the survival function of a lognormal probability density function (PDF) fitted in the measured occupational exposure data. RESULTS: The estimated probabilities of exceeding annual effective dose limits of 1 mSv, 6 mSv, and 20 mSv were in the order of 1:200, 1:10,000, and 1:500,000 per person, respectively. CONCLUSION: It is very unlikely that the Category B annual effective dose limit of 6 mSv could even potentially be exceeded using modern equipment and appropriate working methods. Therefore, in terms of estimated effective dose, workers in diagnostic and interventional radiology could be placed into Category B in Finland. Current national personal monitoring practice could be replaced or supplemented using active personal dosimeters, which offer more effective means for optimizing working methods.

Automated daily quality control analysis for mammography in a multi-unit imaging center.

BACKGROUND: The high requirements for mammography image quality necessitate a systematic quality assurance process. Digital imaging allows automation of the image quality analysis, which can potentially improve repeatability and objectivity compared to a visual evaluation made by the users. PURPOSE: To develop an automatic image quality analysis software for daily mammography quality control in a multi-unit imaging center. MATERIAL AND METHODS: An automated image quality analysis software using the discrete wavelet transform and multiresolution analysis was developed for the American College of Radiology accreditation phantom. The software was validated by analyzing 60 randomly selected phantom images from six mammography systems and 20 phantom images with different dose levels from one mammography system. The results were compared to a visual analysis made by four reviewers. Additionally, long-term image quality trends of a full-field digital mammography system and a computed radiography mammography system were investigated. RESULTS: The automated software produced feature detection levels comparable to visual analysis. The agreement was good in the case of fibers, while the software detected somewhat more microcalcifications and characteristic masses. Long-term follow-up via a quality assurance web portal demonstrated the feasibility of using the software for monitoring the performance of mammography systems in a multi-unit imaging center. CONCLUSION: Automated image quality analysis enables monitoring the performance of digital mammography systems in an efficient, centralized manner.

Data quality in fMRI and simultaneous EEG-fMRI.

OBJECT: To evaluate functional magnetic resonance imaging (fMRI) and simultaneous electroencephalography (EEG)-fMRI data quality in an organization using several magnetic resonance imaging (MRI) systems. MATERIALS AND METHODS: Functional magnetic resonance imaging measurements were carried out twice with a uniform gel phantom on five different MRI systems with field strengths of 1.5 and 3.0 T. Several image quality parameters were measured with automatic analysis software. For simultaneous EEG-fMRI, data quality was evaluated on 3.0 T systems, and the phantom results were compared to data on human volunteers. RESULTS: The fMRI quality parameters measured with different MRI systems were on an acceptable level. The presence of the EEG equipment caused superficial artifacts on the phantom image. The typical artifact depth was 15 mm, and no artifacts were observed in the brain area in the images of volunteers. Average signal-to-noise ratio (SNR) reduction in the phantom measurements was 15 %, a reduction of SNR similar to that observed in the human data. We also detected minor changes in the noise of the EEG signal during the phantom measurement. CONCLUSION: The phantom proved valuable in the successful evaluation of the data quality of fMRI and EEG-fMRI. The results fell within acceptable limits. This study demonstrated a repeatable method to measure and follow up on the data quality of simultaneous EEG-fMRI.

Factors influencing the reliability of a CT angiography-based deep learning method for infarct volume estimation.

Objectives: CT angiography (CTA)-based machine learning methods for infarct volume estimation have shown a tendency to overestimate infarct core and final infarct volumes (FIV). Our aim was to assess factors influencing the reliability of these methods. Methods: The effect of collateral circulation on the correlation between convolutional neural network (CNN) estimations and FIV was assessed based on the Miteff system and hypoperfusion intensity ratio (HIR) in 121 patients with anterior circulation acute ischaemic stroke using Pearson correlation coefficients and median volumes. Correlation was also assessed between successful and futile thrombectomies. The timing of individual CTAs in relation to CTP studies was analysed. Results: The strength of correlation between CNN estimated volumes and FIV did not change significantly depending on collateral status as assessed with the Miteff system or HIR, being poor to moderate (r = 0.09-0.50). The strongest correlation was found in patients with futile thrombectomies (r = 0.61). Median CNN estimates showed a trend for overestimation compared to FIVs. CTA was acquired in the mid arterial phase in virtually all patients (120/121). Conclusions: This study showed no effect of collateral status on the reliability of the CNN and best correlation was found in patients with futile thrombectomies. CTA timing in the mid arterial phase in virtually all patients can explain infarct volume overestimation. Advances in knowledge: CTA timing seems to be the most important factor influencing the reliability of current CTA-based machine learning methods, emphasizing the need for CTA protocol optimization for infarct core estimation.

Deep learning-aided extraction of outer aortic surface from CT angiography scans of patients with Stanford type B aortic dissection.

BACKGROUND: Guidelines recommend that aortic dimension measurements in aortic dissection should include the aortic wall. This study aimed to evaluate two-dimensional (2D)- and three-dimensional (3D)-based deep learning approaches for extraction of outer aortic surface in computed tomography angiography (CTA) scans of Stanford type B aortic dissection (TBAD) patients and assess the speed of different whole aorta (WA) segmentation approaches. METHODS: A total of 240 patients diagnosed with TBAD between January 2007 and December 2019 were retrospectively reviewed for this study; 206 CTA scans from 206 patients with acute, subacute, or chronic TBAD acquired with various scanners in multiple different hospital units were included. Ground truth (GT) WAs for 80 scans were segmented by a radiologist using an open-source software. The remaining 126 GT WAs were generated via semi-automatic segmentation process in which an ensemble of 3D convolutional neural networks (CNNs) aided the radiologist. Using 136 scans for training, 30 for validation, and 40 for testing, 2D and 3D CNNs were trained to automatically segment WA. Main evaluation metrics for outer surface extraction and segmentation accuracy were normalized surface Dice (NSD) and Dice coefficient score (DCS), respectively. RESULTS: 2D CNN outperformed 3D CNN in NSD score (0.92 versus 0.90, p = 0.009), and both CNNs had equal DCS (0.96 versus 0.96, p = 0.110). Manual and semi-automatic segmentation times of one CTA scan were approximately 1 and 0.5 h, respectively. CONCLUSIONS: Both CNNs segmented WA with high DCS, but based on NSD, better accuracy may be required before clinical application. CNN-based semi-automatic segmentation methods can expedite the generation of GTs. RELEVANCE STATEMENT: Deep learning can speeds up the creation of ground truth segmentations. CNNs can extract the outer aortic surface in patients with type B aortic dissection. KEY POINTS: • 2D and 3D convolutional neural networks (CNNs) can extract the outer aortic surface accurately. • Equal Dice coefficient score (0.96) was reached with 2D and 3D CNNs. • Deep learning can expedite the creation of ground truth segmentations.

Leveraging open dataset and transfer learning for accurate recognition of chronic pulmonary embolism from CT angiogram maximum intensity projection images.

BACKGROUND: Early diagnosis of the potentially fatal but curable chronic pulmonary embolism (CPE) is challenging. We have developed and investigated a novel convolutional neural network (CNN) model to recognise CPE from CT pulmonary angiograms (CTPA) based on the general vascular morphology in two-dimensional (2D) maximum intensity projection images. METHODS: A CNN model was trained on a curated subset of a public pulmonary embolism CT dataset (RSPECT) with 755 CTPA studies, including patient-level labels of CPE, acute pulmonary embolism (APE), or no pulmonary embolism. CPE patients with right-to-left-ventricular ratio (RV/LV) < 1 and APE patients with RV/LV ≥ 1 were excluded from the training. Additional CNN model selection and testing were done on local data with 78 patients without the RV/LV-based exclusion. We calculated area under the receiver operating characteristic curves (AUC) and balanced accuracies to evaluate the CNN performance. RESULTS: We achieved a very high CPE versus no-CPE classification AUC 0.94 and balanced accuracy 0.89 on the local dataset using an ensemble model and considering CPE to be present in either one or both lungs. CONCLUSIONS: We propose a novel CNN model with excellent predictive accuracy to differentiate chronic pulmonary embolism with RV/LV ≥ 1 from acute pulmonary embolism and non-embolic cases from 2D maximum intensity projection reconstructions of CTPA. RELEVANCE STATEMENT: A DL CNN model identifies chronic pulmonary embolism from CTA with an excellent predictive accuracy. KEY POINTS: • Automatic recognition of CPE from computed tomography pulmonary angiography was developed. • Deep learning was applied on two-dimensional maximum intensity projection images. • A large public dataset was used for training the deep learning model. • The proposed model showed an excellent predictive accuracy.

Accuracy of the electron transport in mcnp5 and its suitability for ionization chamber response simulations: A comparison with the egsnrc and penelope codes.

PURPOSE: In this work, accuracy of the mcnp5 code in the electron transport calculations and its suitability for ionization chamber (IC) response simulations in photon beams are studied in comparison to egsnrc and penelope codes. METHODS: The electron transport is studied by comparing the depth dose distributions in a water phantom subdivided into thin layers using incident energies (0.05, 0.1, 1, and 10 MeV) for the broad parallel electron beams. The IC response simulations are studied in water phantom in three dosimetric gas materials (air, argon, and methane based tissue equivalent gas) for photon beams ((60)Co source, 6 MV linear medical accelerator, and mono-energetic 2 MeV photon source). Two optional electron transport models of mcnp5 are evaluated: the ITS-based electron energy indexing (mcnp5(ITS)) and the new detailed electron energy-loss straggling logic (mcnp5(new)). The electron substep length (ESTEP parameter) dependency in mcnp5 is investigated as well. RESULTS: For the electron beam studies, large discrepancies (>3%) are observed between the MCNP5 dose distributions and the reference codes at 1 MeV and lower energies. The discrepancy is especially notable for 0.1 and 0.05 MeV electron beams. The boundary crossing artifacts, which are well known for the mcnp5(ITS), are observed for the mcnp5(new) only at 0.1 and 0.05 MeV beam energies. If the excessive boundary crossing is eliminated by using single scoring cells, the mcnp5(ITS) provides dose distributions that agree better with the reference codes than mcnp5(new). The mcnp5 dose estimates for the gas cavity agree within 1% with the reference codes, if the mcnp5(ITS) is applied or electron substep length is set adequately for the gas in the cavity using the mcnp5(new). The mcnp5(new) results are found highly dependent on the chosen electron substep length and might lead up to 15% underestimation of the absorbed dose. CONCLUSIONS: Since the mcnp5 electron transport calculations are not accurate at all energies and in every medium by general clinical standards, caution is needed, if mcnp5 is used with the current electron transport models for dosimetric applications.

Detection of Aspartylglucosaminuria Patients from Magnetic Resonance Images by a Machine-Learning-Based Approach.

Magnetic resonance (MR) imaging data can be used to develop computer-assisted diagnostic tools for neurodegenerative diseases such as aspartylglucosaminuria (AGU) and other lysosomal storage disorders. MR images contain features that are suitable for the classification and differentiation of affected individuals from healthy persons. Here, comparisons were made between MRI features extracted from different types of magnetic resonance images. Random forest classifiers were trained to classify AGU patients (n = 22) and healthy controls (n = 24) using volumetric features extracted from T1-weighted MR images, the zone variance of gray level size zone matrix (GLSZM) calculated from magnitude susceptibility-weighted MR images, and the caudate-thalamus intensity ratio computed from T2-weighted MR images. The leave-one-out cross-validation and area under the receiver operating characteristic curve were used to compare different models. The left-right-averaged, normalized volumes of the 25 nuclei of the thalamus and the zone variance of the thalamus demonstrated equal and excellent performance as classifier features for binary organization between AGU patients and healthy controls. Our findings show that texture-based features of susceptibility-weighted images and thalamic volumes can differentiate AGU patients from healthy controls with a very low error rate.

Stiffness-Controlled Hydrogels for 3D Cell Culture Models.

Nanofibrillated cellulose (NFC) hydrogel is a versatile biomaterial suitable, for example, for three-dimensional (3D) cell spheroid culturing, drug delivery, and wound treatment. By freeze-drying NFC hydrogel, highly porous NFC structures can be manufactured. We freeze-dried NFC hydrogel and subsequently reconstituted the samples into a variety of concentrations of NFC fibers, which resulted in different stiffness of the material, i.e., different mechanical cues. After the successful freeze-drying and reconstitution, we showed that freeze-dried NFC hydrogel can be used for one-step 3D cell spheroid culturing of primary mesenchymal stem/stromal cells, prostate cancer cells (PC3), and hepatocellular carcinoma cells (HepG2). No difference was observed in the viability or morphology between the 3D cell spheroids cultured in the freeze-dried and reconstituted NFC hydrogel and fresh NFC hydrogel. Furthermore, the 3D cultured spheroids showed stable metabolic activity and nearly 100% viability. Finally, we applied a convolutional neural network (CNN)-based automatic nuclei segmentation approach to automatically segment individual cells of 3D cultured PC3 and HepG2 spheroids. These results provide an application to culture 3D cell spheroids more readily with the NFC hydrogel and a step towards automatization of 3D cell culturing and analysis.

MRI quality assurance using the ACR phantom in a multi-unit imaging center.

BACKGROUND: Magnetic resonance imaging (MRI) instrumentation is vulnerable to technical and image quality problems, and quality assurance is essential. In the studied regional imaging center the long-term quality assurance has been based on MagNET phantom measurements. American College of Radiology (ACR) has an accreditation program including a standardized image quality measurement protocol and phantom. The ACR protocol includes recommended acceptance criteria for clinical sequences and thus provides possibility to assess the clinical relevance of quality assurance. The purpose of this study was to test the ACR MRI phantom in quality assurance of a multi-unit imaging center. MATERIAL AND METHODS: The imaging center operates 11 MRI systems of three major manufacturers with field strengths of 3.0 T, 1.5 T and 1.0 T. Images of the ACR phantom were acquired using a head coil following the ACR scanning instructions. Both ACR T1- and T2-weighted sequences as well as T1- and T2-weighted brain sequences in clinical use at each site were acquired. Measurements were performed twice. The images were analyzed and the results were compared with the ACR acceptance levels. RESULTS: The acquisition procedure with the ACR phantom was faster than with the MagNET phantoms. On the first and second measurement rounds 91% and 73% of the systems passed the ACR test. Measured slice thickness accuracies were not within the acceptance limits in site T2 sequences. Differences in the high contrast spatial resolution between the ACR and the site sequences were observed. In 3.0 T systems the image intensity uniformity was slightly lower than the ACR acceptance limit. CONCLUSION: The ACR method was feasible in quality assurance of a multi-unit imaging center and the ACR protocol could replace the MagNET phantom tests. An automatic analysis of the images will further improve cost-effectiveness and objectiveness of the ACR protocol.

Computed tomography angiography-based deep learning method for treatment selection and infarct volume prediction in anterior cerebral circulation large vessel occlusion.

BACKGROUND: Computed tomography perfusion (CTP) is the mainstay to determine possible eligibility for endovascular thrombectomy (EVT), but there is still a need for alternative methods in patient triage. PURPOSE: To study the ability of a computed tomography angiography (CTA)-based convolutional neural network (CNN) method in predicting final infarct volume in patients with large vessel occlusion successfully treated with endovascular therapy. MATERIALS AND METHODS: The accuracy of the CTA source image-based CNN in final infarct volume prediction was evaluated against follow-up CT or MR imaging in 89 patients with anterior circulation ischemic stroke successfully treated with EVT as defined by Thrombolysis in Cerebral Infarction category 2b or 3 using Pearson correlation coefficients and intraclass correlation coefficients. Convolutional neural network performance was also compared to a commercially available CTP-based software (RAPID, iSchemaView). RESULTS: A correlation with final infarct volumes was found for both CNN and CTP-RAPID in patients presenting 6-24 h from symptom onset or last known well, with r = 0.67 (p < 0.001) and r = 0.82 (p < 0.001), respectively. Correlations with final infarct volumes in the early time window (0-6 h) were r = 0.43 (p = 0.002) for the CNN and r = 0.58 (p < 0.001) for CTP-RAPID. Compared to CTP-RAPID predictions, CNN estimated eligibility for thrombectomy according to ischemic core size in the late time window with a sensitivity of 0.38 and specificity of 0.89. CONCLUSION: A CTA-based CNN method had moderate correlation with final infarct volumes in the late time window in patients successfully treated with EVT.

Evaluation of a CTA-based convolutional neural network for infarct volume prediction in anterior cerebral circulation ischaemic stroke.

BACKGROUND: Computed tomography angiography (CTA) imaging is needed in current guideline-based stroke diagnosis, and infarct core size is one factor in guiding treatment decisions. We studied the efficacy of a convolutional neural network (CNN) in final infarct volume prediction from CTA and compared the results to a CT perfusion (CTP)-based commercially available software (RAPID, iSchemaView). METHODS: We retrospectively selected 83 consecutive stroke cases treated with thrombolytic therapy or receiving supportive care that presented to Helsinki University Hospital between January 2018 and July 2019. We compared CNN-derived ischaemic lesion volumes to final infarct volumes that were manually segmented from follow-up CT and to CTP-RAPID ischaemic core volumes. RESULTS: An overall correlation of r = 0.83 was found between CNN outputs and final infarct volumes. The strongest correlation was found in a subgroup of patients that presented more than 9 h of symptom onset (r = 0.90). A good correlation was found between the CNN outputs and CTP-RAPID ischaemic core volumes (r = 0.89) and the CNN was able to classify patients for thrombolytic therapy or supportive care with a 1.00 sensitivity and 0.94 specificity. CONCLUSIONS: A CTA-based CNN software can provide good infarct core volume estimates as observed in follow-up imaging studies. CNN-derived infarct volumes had a good correlation to CTP-RAPID ischaemic core volumes.

Performance of a 3D convolutional neural network in the detection of hypoperfusion at CT pulmonary angiography in patients with chronic pulmonary embolism: a feasibility study.

BACKGROUND: Chronic pulmonary embolism (CPE) is a life-threatening disease easily misdiagnosed on computed tomography. We investigated a three-dimensional convolutional neural network (CNN) algorithm for detecting hypoperfusion in CPE from computed tomography pulmonary angiography (CTPA). METHODS: Preoperative CTPA of 25 patients with CPE and 25 without pulmonary embolism were selected. We applied a 48%-12%-40% training-validation-testing split (12 positive and 12 negative CTPA volumes for training, 3 positives and 3 negatives for validation, 10 positives and 10 negatives for testing). The median number of axial images per CTPA was 335 (min-max, 111-570). Expert manual segmentations were used as training and testing targets. The CNN output was compared to a method in which a Hounsfield unit (HU) threshold was used to detect hypoperfusion. Receiver operating characteristic area under the curve (AUC) and Matthew correlation coefficient (MCC) were calculated with their 95% confidence interval (CI). RESULTS: The predicted segmentations of CNN showed AUC 0.87 (95% CI 0.82-0.91), those of HU-threshold method 0.79 (95% CI 0.74-0.84). The optimal global threshold values were CNN output probability ≥ 0.37 and ≤ -850 HU. Using these values, MCC was 0.46 (95% CI 0.29-0.59) for CNN and 0.35 (95% CI 0.18-0.48) for HU-threshold method (average difference in MCC in the bootstrap samples 0.11 (95% CI 0.05-0.16). A high CNN prediction probability was a strong predictor of CPE. CONCLUSIONS: We proposed a deep learning method for detecting hypoperfusion in CPE from CTPA. This model may help evaluating disease extent and supporting treatment planning.

Acquisition-weighted MRSI for detection and quantification of BNCT 10B-carrier L-p-boronophenylalanine-fructose complex, a phantom study.

Proton magnetic resonance spectroscopy (1H MRS) is a potential method to detect and quantify a boron neutron capture therapy 10B-carrier compound, L-p-boronophenylalanine (BPA), in the brain. However, optimal positioning of MRS voxel to capture tissue with maximal BPA concentration can be challenging. Three dimensional proton magnetic resonance spectroscopic imaging (3D 1H MRSI) provides spectral data covering a large spatial volume, which is a major advantage in detecting and quantifying BPA. BPA detection limit in phantom conditions was determined at 1.5 T using a 3D 1H MRSI protocol with clinically acceptable nominal spatial resolution and duration. Quantification tests for aqueous phantom were performed using both single voxel MRS and 3D MRSI. In 3D MRSI, BPA detection limit was approximately 1.0 mM and BPA quantification accuracy was better than +/-5%. The results suggest that MRSI would be a feasible method for in vivo BPA evaluation in clinical conditions.

3D convolutional neural networks applied to CT angiography in the detection of acute ischemic stroke.

BACKGROUND: The aim of this study was to investigate the feasibility of ischemic stroke detection from computed tomography angiography source images (CTA-SI) using three-dimensional convolutional neural networks. METHODS: CTA-SI of 60 patients with a suspected acute ischemic stroke of the middle cerebral artery were randomly selected for this study; 30 patients were used in the neural network training, and the subsequent testing was performed using the remaining 30 patients. The training and testing were based on manually segmented lesions. Cerebral hemispheric comparison CTA and non-contrast computed tomography (NCCT) were studied as additional input features. RESULTS: All ischemic lesions in the testing data were correctly lateralized, and a high correspondence to manual segmentations was achieved. Patients with a diagnosed stroke had clinically relevant regions labeled infarcted with a 0.93 sensitivity and 0.82 specificity. The highest achieved voxel-wise area under receiver operating characteristic curve was 0.93, and the highest Dice similarity coefficient was 0.61. When cerebral hemispheric comparison was used as an input feature, the algorithm performance improved. Only a slight effect was seen when NCCT was included. CONCLUSION: The results support the hypothesis that an acute ischemic stroke lesion can be detected with 3D convolutional neural network-based software from CTA-SI. Utilizing information from the contralateral hemisphere appears to be beneficial for reducing false positive findings.

Boron neutron capture therapy in the treatment of locally recurred head and neck cancer.

PURPOSE: Head and neck carcinomas that recur locally after conventional irradiation pose a difficult therapeutic problem. We evaluated safety and efficacy of boron neutron capture therapy (BNCT) in the treatment of such cancers. METHODS AND MATERIALS: Twelve patients with inoperable, recurred, locally advanced (rT3, rT4, or rN2) head and neck cancer were treated with BNCT in a prospective, single-center Phase I-II study. Prior treatments consisted of surgery and conventionally fractionated photon irradiation to a cumulative dose of 56-74 Gy administered with or without concomitant chemotherapy. Tumor responses were assessed using the RECIST (Response Evaluation Criteria in Solid Tumors) criteria and adverse effects using the National Cancer Institute common toxicity grading v3.0. Intravenously administered boronophenylalanine-fructose (BPA-F, 400 mg/kg) was used as the boron carrier. Each patient was scheduled to be treated twice with BNCT. RESULTS: Ten patients received BNCT twice; 2 were treated once. Ten (83%) patients responded to BNCT, and 2 (17%) had tumor growth stabilization for 5.5 and 7.6 months. The median duration of response was 12.1 months; six responses were ongoing at the time of analysis or death (range, 4.9-19.2 months). Four (33%) patients were alive without recurrence with a median follow-up of 14.0 months (range, 12.8-19.2 months). The most common acute adverse effects were mucositis, fatigue, and local pain; 2 patients had a severe (Grade 3) late adverse effect (xerostomia, 1; dysphagia, 1). CONCLUSIONS: Boron neutron capture therapy is effective and safe in the treatment of inoperable, locally advanced head and neck carcinomas that recur at previously irradiated sites.