Organ Contouring for Lung Cancer Patients with a Seed Generation Scheme and Random Walks.

In this study, we proposed a semi-automated and interactive scheme for organ contouring in radiotherapy planning for patients with non-small cell lung cancers. Several organs were contoured, including the lungs, airway, heart, spinal cord, body, and gross tumor volume (GTV). We proposed some schemes to automatically generate and vanish the seeds of the random walks (RW) algorithm. We considered 25 lung cancer patients, whose computed tomography (CT) images were obtained from the China Medical University Hospital (CMUH) in Taichung, Taiwan. The manual contours made by clinical oncologists were taken as the gold standard for comparison to evaluate the performance of our proposed method. The Dice coefficient between two contours of the same organ was computed to evaluate the similarity. The average Dice coefficients for the lungs, airway, heart, spinal cord, and body and GTV segmentation were 0.92, 0.84, 0.83, 0.73, 0.85 and 0.66, respectively. The computation time was between 2 to 4 min for a whole CT sequence segmentation. The results showed that our method has the potential to assist oncologists in the process of radiotherapy treatment in the CMUH, and hopefully in other hospitals as well, by saving a tremendous amount of time in contouring.

Noninvasive assessment of intracranial elastance and pressure in spontaneous intracranial hypotension by MRI.

BACKGROUND: Spontaneous intracranial hypotension (SIH) is often misdiagnosed, and can lead to severe complications. Conventional MR sequences show a limited ability to aid in this diagnosis. MR-based intracranial pressure (MR-ICP) may be able to detect changes of intracranial elastance and pressure. PURPOSE: To determine whether MR-ICP is able to differentiate SIH patients from normal subjects, improve diagnostic sensitivity, and provide an insight into the pathophysiology. STUDY TYPE: Prospective. SUBJECTS: Twenty-eight SIH cases with orthostatic headache and 20 healthy volunteers. FIELD STRENGTH/SEQUENCE: Cine phase-contrast MRI on a 1.5T scanner. ASSESSMENT: Intracranial elastance (IE) was derived from the ratio of the peak-to-peak cerebrospinal fluid (CSF) pressure gradient (PGcsf-pp ) and intracranial volume change, obtained by summing all flows before each sequential cardiac frame. STATISTICAL TESTS: Student's t-test was used to compare the MR-ICP indexes and flow parameters between SIH patients and healthy volunteers (P < 0.01). RESULTS: The SIH patients with cervical epidural venous dilatation (EVD) had an IE of 0.121 ± 0.027 mmHg/cm/ml, significantly higher than that of the normal volunteers (0.085 ± 0.027 mmHg/cm/ml; P = 0.002). In contradistinction, the EVD-negative SIH patients, including four with no sign of CSF leaks, had significantly lower IE (0.055 ± 0.012 mmHg/cm/ml) compared with the normal volunteers and the EVD-positive group (P = 0.001, P < 0.001). The EVD-negative patients had significantly lower PGcsf-pp (0.024 ± 0.007 mmHg/cm) compared with the normal volunteers and the EVD-positive group (0.035 ± 0.011 mmHg/cm, 0.040 ± 0.010 mmHg/cm; P = 0.003, P < 0.001). Additionally, the MRI flow study showed a significant decrease in transcranial inflow and outflow of SIH patients (P < 0.01). DATA CONCLUSION: We found that the MR-ICP method is potentially more sensitive than morphological MRI in the early diagnosis of SIH. Also, contrary to common belief, our results suggest that an abnormal craniospinal elastance might be the cause of SIH, instead of CSF leak. LEVEL OF EVIDENCE: 2 Technical Efficacy: Stage 3 J. Magn. Reson. Imaging 2018;47:1255-1263.

Improvement of Left Ventricular Ejection Time Measurement in the Impedance Cardiography Combined with the Reflection Photoplethysmography.

Cardiac stroke volume (SV) is an essential hemodynamic indicator that can be used to assess whether the pump function of the heart is normal. Non-invasive SV measurement is currently performed using the impedance cardiography (ICG). In this technology, left ventricular ejection time (LVET) is an important parameter which can be determined from the ICG signals. However, the ICG signals are inherently susceptible to artificial noise interference, which leads to an inaccurate LVET measurement and then yields an error in the calculation of SV. Therefore, the goal of the study was to measure LVETs using both the transmission and reflection photoplethysmography (PPG), and to assess whether the measured LVET was more accurate by the PPG signal than the ICG signal. The LVET measured by the phonocardiography (PCG) was used as the standard for comparing with those by the ICG and PPG. The study recruited ten subjects whose LVETs were simultaneously measured by the ICG using four electrodes, the reflection PPG using neck sensors (PPGneck) and the transmission PPG using finger sensors (PPGfinger). In each subject, ten LVETs were obtained from ten heartbeats selected properly from one-minute recording. The differences of the measured LVETs between the PCG and one of the ICG, PPGneck and PPGfinger were -68.2 ± 148.6 ms, 4.8 ± 86.5 ms and -7.0 ± 107.5 ms, respectively. As compared with the PCG, both the ICG and PPGfinger underestimated but the PPGneck overestimated the LVETs. Furthermore, the measured LVET by the PPGneck was the closest to that by the PCG. Therefore, the PPGneck may be employed to improve the LVET measurement in applying the ICG for continuous monitoring of SV in clinical settings.

A Cuffless Blood Pressure Measurement Based on the Impedance Plethysmography Technique.

In the last decade, cuffless blood pressure measurement technology has been widely studied because it could be applied to a wearable apparatus. Electrocardiography (ECG), photo-plethysmography (PPG), and phonocardiography are always used to detect the pulse transit time (PTT) because the changed tendencies of the PTT and blood pressure have a negative relationship. In this study, the PPG signal was replaced by the impedance plethysmography (IPG) signal and was used to detect the PTT. The placement and direction of the electrode array for the IPG measurement were discussed. Then, we designed an IPG ring that could measure an accurate IPG signal. Twenty healthy subjects participated in this study. The changes in blood pressure after exercise were evaluated through the changes of the PTT. The results showed that the change of the systolic pressure had a better relationship with the change of the PTTIPG than that of the PTTPPG (r = 0.700 vs. r = 0.450). Moreover, the IPG ring with spot electrodes would be more suitable to develop with the wearable cuffless blood pressure monitor than the PPG sensor.

Assessment of the endothelial function with changed volume of brachial artery by menstrual cycle.

BACKGROUND: The endothelial function has been proven to be an important factor in the pathogenesis of atherosclerosis, hypertension and heart failure. The flow-mediated vasodilation (FMD) of the peripheral artery is an endothelium-dependent function. Brachial-artery ultrasound scanning is the popular method for evaluating FMD. However, good technical training on ultrasonography is required for the user to obtain high-quality data. Therefore, the goal of this study was to propose a new method which only used a sphygmomanometer cuff to occlude the blood flow and record the vascular volume waveform (Vwave). RESULTS: We used this method to assess the FMD in the menstrual cycle for 26 volunteer females. All female subjects were evaluated two times (M: menstrual phase; F: luteal phase) in one menstrual cycle and for two cycles. In the first cycle, the FMD volume ratio in M was 101.9 ± 45.5 % and was higher in L, at 137.5 ± 62.1 % (p = 0.0032 versus M). In the second cycle, the FMD volume ratios in M and L were 91.4 ± 37.0 % and 124.0 ± 56.4 %, respectively (p < 0.001 vs. M). CONCLUSIONS: Our results have confirmed those results in the study of Hametner et al. Blood pressure measurement and FMD assessment all used the same mechanic of digital blood pressure monitor, which makes our method suitable using at home.

Accurate Measurement of Cross-Sectional Area of Femoral Artery on MRI Sequences of Transcontinental Ultramarathon Runners Using Optimal Parameters Selection.

In clinics an accurate vessel segmentation method is important to quantize the vessel volume change with respect to time for artery elasticity measurement. This study proposes a modified version on 3D-expanded dynamic programming to find an optimal surface in a 3D matrix. The aim of this study is to discover the robustness against noises in measuring the cross-sectional area of the femoral artery on MRI datasets of ultra-endurance runners as accurately as possible. To do this, we use phantom images with different added noises and different image contrasts to find out the optimal parameters using grid search. The contrast between the vessel lumen and its background in phantom study is changed to simulate the real MRI dataset. We also add a plaque in phantom images to test the accuracy of the proposed algorithm in dealing pathologic cases. The phantom studies and grid search on selecting optimal parameters can offer an alternative way on parameter selection. In application to MRI, the accuracy is performed via comparisons between the manual tracings of experts and automated results. The mean relative error is 2.1 % ± 2.1 % on testing 11 MRI datasets (total 550 images). The phantom studies and grid search on selecting optimal parameters can offer an alternative way on parameter selection.

Remote Diagnosis on Upper Respiratory Tract Infections Based on a Neural Network with Few Symptom Words-A Feasibility Study.

This study aims explore the feasibility of using neural network (NNs) and deep learning to diagnose three common respiratory diseases with few symptom words. These three diseases are nasopharyngitis, upper respiratory infection, and bronchitis/bronchiolitis. Through natural language processing, the symptom word vectors are encoded by GPT-2 and classified by the last linear layer of the NN. The experimental results are promising, showing that this model achieves a high performance in predicting all three diseases. They revealed 90% accuracy, which suggests the implications of the developed model, highlighting its potential use in assisting patients' understanding of their conditions via a remote diagnosis. Unlike previous studies that have focused on extracting various categories of information from medical records, this study directly extracts sequential features from unstructured text data, reducing the effort required for data pre-processing.

Arrhythmia identification with two-lead electrocardiograms using artificial neural networks and support vector machines for a portable ECG monitor system.

An automatic configuration that can detect the position of R-waves, classify the normal sinus rhythm (NSR) and other four arrhythmic types from the continuous ECG signals obtained from the MIT-BIH arrhythmia database is proposed. In this configuration, a support vector machine (SVM) was used to detect and mark the ECG heartbeats with raw signals and differential signals of a lead ECG. An algorithm based on the extracted markers segments waveforms of Lead II and V1 of the ECG as the pattern classification features. A self-constructing neural fuzzy inference network (SoNFIN) was used to classify NSR and four arrhythmia types, including premature ventricular contraction (PVC), premature atrium contraction (PAC), left bundle branch block (LBBB), and right bundle branch block (RBBB). In a real scenario, the classification results show the accuracy achieved is 96.4%. This performance is suitable for a portable ECG monitor system for home care purposes.

Automated feature set selection and its application to MCC identification in digital mammograms for breast cancer detection.

We propose a fully automated algorithm that is able to select a discriminative feature set from a training database via sequential forward selection (SFS), sequential backward selection (SBS), and F-score methods. We applied this scheme to microcalcifications cluster (MCC) detection in digital mammograms for early breast cancer detection. The system was able to select features fully automatically, regardless of the input training mammograms used. We tested the proposed scheme using a database of 111 clinical mammograms containing 1,050 microcalcifications (MCs). The accuracy of the system was examined via a free response receiver operating characteristic (fROC) curve of the test dataset. The system performance for MC identifications was Az = 0.9897, the sensitivity was 92%, and 0.65 false positives (FPs) were generated per image for MCC detection.

Skeleton Segmentation on Bone Scintigraphy for BSI Computation.

Bone Scan Index (BSI) is an image biomarker for quantifying bone metastasis of cancers. To compute BSI, not only the hotspots (metastasis) but also the bones have to be segmented. Most related research focus on binary classification in bone scintigraphy: having metastasis or none. Rare studies focus on pixel-wise segmentation. This study compares three advanced convolutional neural network (CNN) based models to explore bone segmentation on a dataset in-house. The best model is Mask R-CNN, which reaches the precision, sensitivity, and F1-score: 0.93, 0.87, 0.90 for prostate cancer patients and 0.92, 0.86, and 0.88 for breast cancer patients, respectively. The results are the average of 10-fold cross-validation, which reveals the reliability of clinical use on bone segmentation.

Impact of Voxel Normalization on a Machine Learning-Based Method: A Study on Pulmonary Nodule Malignancy Diagnosis Using Low-Dose Computed Tomography (LDCT).

Lung cancer (LC) stands as the foremost cause of cancer-related fatality rates worldwide. Early diagnosis significantly enhances patient survival rate. Nowadays, low-dose computed tomography (LDCT) is widely employed on the chest as a tool for large-scale lung cancer screening. Nonetheless, a large amount of chest radiographs creates an onerous burden for radiologists. Some computer-aided diagnostic (CAD) tools can provide insight to the use of medical images for diagnosis and can augment diagnostic speed. However, due to the variation in the parameter settings across different patients, substantial discrepancies in image voxels persist. We found that different voxel sizes can create a compromise between model generalization and diagnostic efficacy. This study investigates the performance disparities of diagnostic models trained on original images and LDCT images reconstructed to different voxel sizes while making isotropic. We examined the ability of our method to differentiate between benign and malignant nodules. Using 11 features, a support vector machine (SVM) was trained on LDCT images using an isotropic voxel with a side length of 1.5 mm for 225 patients in-house. The result yields a favorable model performance with an accuracy of 0.9596 and an area under the receiver operating characteristic curve (ROC/AUC) of 0.9855. In addition, to furnish CAD tools for clinical application, future research including LDCT images from multi-centers is encouraged.

Bone Metastases Lesion Segmentation on Breast Cancer Bone Scan Images with Negative Sample Training.

The use of deep learning methods for the automatic detection and quantification of bone metastases in bone scan images holds significant clinical value. A fast and accurate automated system for segmenting bone metastatic lesions can assist clinical physicians in diagnosis. In this study, a small internal dataset comprising 100 breast cancer patients (90 cases of bone metastasis and 10 cases of non-metastasis) and 100 prostate cancer patients (50 cases of bone metastasis and 50 cases of non-metastasis) was used for model training. Initially, all image labels were binary. We used the Otsu thresholding method or negative mining to generate a non-metastasis mask, thereby transforming the image labels into three classes. We adopted the Double U-Net as the baseline model and made modifications to its output activation function. We changed the activation function to SoftMax to accommodate multi-class segmentation. Several methods were used to enhance model performance, including background pre-processing to remove background information, adding negative samples to improve model precision, and using transfer learning to leverage shared features between two datasets, which enhances the model's performance. The performance was investigated via 10-fold cross-validation and computed on a pixel-level scale. The best model we achieved had a precision of 69.96%, a sensitivity of 63.55%, and an F1-score of 66.60%. Compared to the baseline model, this represents an 8.40% improvement in precision, a 0.56% improvement in sensitivity, and a 4.33% improvement in the F1-score. The developed system has the potential to provide pre-diagnostic reports for physicians in final decisions and the calculation of the bone scan index (BSI) with the combination with bone skeleton segmentation.

Three-dimensional expansion of a dynamic programming method for boundary detection and its application to sequential magnetic resonance imaging (MRI).

This study proposes a fast 3D dynamic programming expansion to find a shortest surface in a 3D matrix. This algorithm can detect boundaries in an image sequence. Using phantom image studies with added uniform distributed noise from different SNRs, the unsigned error of this proposed method is investigated. Comparing the automated results to the gold standard, the best averaged relative unsigned error of the proposed method is 0.77% (SNR = 20 dB), and its corresponding parameter values are reported. We further apply this method to detect the boundary of the real superficial femoral artery (SFA) in MRI sequences without a contrast injection. The manual tracings on the SFA boundaries are performed by well-trained experts to be the gold standard. The comparisons between the manual tracings and automated results are made on 16 MRI sequences (800 total images). The average unsigned error rate is 2.4% (SD = 2.0%). The results demonstrate that the proposed method can perform qualitatively better than the 2D dynamic programming for vessel boundary detection on MRI sequences.

Efficacy analysis of new copper complex for visible light (455, 530 nm) radical/cationic photopolymerization: The synergic effects and catalytic cycle.

The kinetics and the conversion features of two 3-component systems (A/B/N), based on the proposed new kinetic schemes of Mokbel and Mau et al, in which a visible LED is used to excite a copper complex to its excited triplet state (G*). The coupling of G* with iodonium salt and ethyl 4-(dimethylamino)benzoate (EDB) produces both free radical polymerization (FRP) of acrylates and the free radical promoted cationic polymerization (CP) of epoxides using various new copper complex as the initiator. Higher FRP and CP conversion can be achieved by co-additive of [B] and N, via the dual function of (i) regeneration [A], and (ii) generation of extra radicals. The interpenetrated polymer network (IPN) capable of initiating both FRP and CP in a blend of TMPTA and EPOX. The synergic effects due to CP include: (i) CP can increase viscosity limiting the diffusional oxygen replenishment; (ii) the cation also acts as a diluting agent for the IPN network, and (iii) the exothermic property of the CP. The catalytic cycle, synergic effects, and the oxygen inhibition are theoretically confirmed to support the experimental hypothesis. The measured results of Mokbel and Mau et al are well analyzed and matching the predicted features of our modeling.

Neck Lymph Node Recurrence in HNC Patients Might Be Predicted before Radiotherapy Using Radiomics Extracted from CT Images and XGBoost Algorithm.

The five-year overall survival rate of patients without neck lymph node recurrence is over 50% higher than those with lymph node metastasis. This study aims to investigate the prognostic impact of computed tomogram (CT)-based radiomics on the outcome of metastatic neck lymph nodes in patients with head and neck cancer (HNC) receiving definitive radiotherapy or chemoradiotherapy for organ preservation. The pretreatment 18F-FDG PET/CT of 79 HNC patients was retrospectively analyzed with radiomics extractors. The imbalanced data was processed using two techniques: over-sampling and under-sampling, after which the prediction model was established with a machine learning model using the XGBoost algorithm. The imbalanced dataset strategies slightly decreased the specificity but greatly improved the sensitivity. To have a higher chance of predicting neck cancer recurrence, however, clinical data combined with CT-based radiomics provides the best prediction effect. The original dataset performed was as follows: accuracy = 0.76 ± 0.07, sensitivity = 0.44 ± 0.22, specificity = 0.88 ± 0.06. After we used the over-sampling technique, the accuracy, sensitivity, and specificity values were 0.80 ± 0.05, 0.67 ± 0.11, and 0.84 ± 0.05, respectively. Furthermore, after using the under-sampling technique, the accuracy, sensitivity, and specificity values were 0.71 ± 0.09, 0.73 ± 0.13, and 0.70 ± 0.13, respectively. The outcome of metastatic neck lymph nodes in patients with HNC receiving radiotherapy for organ preservation can be predicted based on the results of machine learning. This way, patients can be treated alternatively. A further external validation study is required to verify our findings.

Prediction of All-Cause Mortality Based on Stress/Rest Myocardial Perfusion Imaging (MPI) Using Deep Learning: A Comparison between Image and Frequency Spectra as Input.

BACKGROUND: Cardiovascular management and risk stratification of patients is an important issue in clinics. Patients who have experienced an adverse cardiac event are concerned for their future and want to know the survival probability. METHODS: We trained eight state-of-the-art CNN models using polar maps of myocardial perfusion imaging (MPI), gender, lung/heart ratio, and patient age for 5-year survival prediction after an adverse cardiac event based on a cohort of 862 patients who had experienced adverse cardiac events and stress/rest MPIs. The CNN model outcome is to predict a patient's survival 5 years after a cardiac event, i.e., two classes, either yes or no. RESULTS: The best accuracy of all the CNN prediction models was 0.70 (median value), which resulted from ResNet-50V2, using image as the input in the baseline experiment. All the CNN models had better performance after using frequency spectra as the input. The accuracy increment was about 7~9%. CONCLUSIONS: This is the first trial to use pure rest/stress MPI polar maps and limited clinical data to predict patients' 5-year survival based on CNN models and deep learning. The study shows the feasibility of using frequency spectra rather than images, which might increase the performance of CNNs.

Modeling the Enhanced Efficacy and Curing Depth of Photo-Thermal Dual Polymerization in Metal (Fe) Polymer Composites for 3D Printing.

This article presents, for the first time, the efficacy and curing depth analysis of photo-thermal dual polymerization in metal (Fe) polymer composites for 3D printing of a three-component (A/B/M) system based on the proposed mechanism of our group, in which the co initiators A and B are Irgacure-369 and charge-transfer complexes (CTC), respectively, and the monomer M is filled by Fe. Our formulas show the depth of curing (Zc) is an increasing function of the light intensity, but a decreasing function of the Fe and photoinitiator concentrations. Zc is enhanced by the additive [B], which produces extra thermal radical for polymerization under high temperature. The heat (or temperature) increase in the system has two components: (i) due to the light absorption of Fe filler and (ii) heat released from the exothermic photopolymerization of the monomer. The heat is transported to the additive (or co-initiator) [B] to produce extra radicals and enhance the monomer conversion function (CF). The Fe filler leads to a temperature increase but also limits the light penetration, leading to lower CF and Zc, which could be overcome by the additive initiator [B] in thick polymers. Optimal Fe for maximal CF and Zc are explored theoretically. Measured data are analyzed based on our derived formulas.

Automatic detection of the carotid artery boundary on cross-sectional MR image sequences using a circle model guided dynamic programming.

BACKGROUND: Systematic aerobe training has positive effects on the compliance of dedicated arterial walls. The adaptations of the arterial structure and function are associated with the blood flow-induced changes of the wall shear stress which induced vascular remodelling via nitric oxide delivered from the endothelial cell. In order to assess functional changes of the common carotid artery over time in these processes, a precise measurement technique is necessary. Before this study, a reliable, precise, and quick method to perform this work is not present. METHODS: We propose a fully automated algorithm to analyze the cross-sectional area of the carotid artery in MR image sequences. It contains two phases: (1) position detection of the carotid artery, (2) accurate boundary identification of the carotid artery. In the first phase, we use intensity, area size and shape as features to discriminate the carotid artery from other tissues and vessels. In the second phase, the directional gradient, Hough transform, and circle model guided dynamic programming are used to identify the boundary accurately. RESULTS: We test the system stability using contrast degraded images (contrast resolutions range from 50% to 90%). The unsigned error ranges from 2.86% ± 2.24% to 3.03% ± 2.40%. The test of noise degraded images (SNRs range from 16 to 20 dB) shows the unsigned error ranging from 2.63% ± 2.06% to 3.12% ± 2.11%. The test of raw images has an unsigned error 2.56% ± 2.10% compared to the manual tracings. CONCLUSIONS: We have proposed an automated system which is able to detect carotid artery cross sectional boundary in MRI sequences during heart cycles. The accuracy reaches 2.56% ± 2.10% compared to the manual tracings. The system is stable, reliable and results are reproducible.

Lesion-Based Bone Metastasis Detection in Chest Bone Scintigraphy Images of Prostate Cancer Patients Using Pre-Train, Negative Mining, and Deep Learning.

This study aimed to explore efficient ways to diagnose bone metastasis early using bone scintigraphy images through negative mining, pre-training, the convolutional neural network, and deep learning. We studied 205 prostate cancer patients and 371 breast cancer patients and used bone scintigraphy data from breast cancer patients to pre-train a YOLO v4 with a false-positive reduction strategy. With the pre-trained model, transferred learning was applied to prostate cancer patients to build a model to detect and identify metastasis locations using bone scintigraphy. Ten-fold cross validation was conducted. The mean sensitivity and precision rates for bone metastasis location detection and classification (lesion-based) in the chests of prostate patients were 0.72 ± 0.04 and 0.90 ± 0.04, respectively. The mean sensitivity and specificity rates for bone metastasis classification (patient-based) in the chests of prostate patients were 0.94 ± 0.09 and 0.92 ± 0.09, respectively. The developed system has the potential to provide pre-diagnostic reports to aid in physicians' final decisions.

Automated detection of the arterial inner walls of the common carotid artery based on dynamic B-mode signals.

In this paper we propose a novel scheme able to automatically detect the intima and adventitia of both near and far walls of the common carotid artery in dynamic B-mode RF (radiofrequency) image sequences, with and without plaques. Via this automated system the lumen diameter changes along the heart cycle can be detected. Three image sequences have been tested and all results are compared to manual tracings made by two professional experts. The average errors for near and far wall detection are 0.058 mm and 0.067 mm, respectively. This system is able to analyze arterial plaques dynamically which is impossible to do manually due to the tremendous human workload involved.