A Multichannel CT and Radiomics-Guided CNN-ViT (RadCT-CNNViT) Ensemble Network for Diagnosis of Pulmonary Sarcoidosis.

Pulmonary sarcoidosis is a multisystem granulomatous interstitial lung disease (ILD) with a variable presentation and prognosis. The early accurate detection of pulmonary sarcoidosis may prevent progression to pulmonary fibrosis, a serious and potentially life-threatening form of the disease. However, the lack of a gold-standard diagnostic test and specific radiographic findings poses challenges in diagnosing pulmonary sarcoidosis. Chest computed tomography (CT) imaging is commonly used but requires expert, chest-trained radiologists to differentiate pulmonary sarcoidosis from lung malignancies, infections, and other ILDs. In this work, we develop a multichannel, CT and radiomics-guided ensemble network (RadCT-CNNViT) with visual explainability for pulmonary sarcoidosis vs. lung cancer (LCa) classification using chest CT images. We leverage CT and hand-crafted radiomics features as input channels, and a 3D convolutional neural network (CNN) and vision transformer (ViT) ensemble network for feature extraction and fusion before a classification head. The 3D CNN sub-network captures the localized spatial information of lesions, while the ViT sub-network captures long-range, global dependencies between features. Through multichannel input and feature fusion, our model achieves the highest performance with accuracy, sensitivity, specificity, precision, F1-score, and combined AUC of 0.93 ± 0.04, 0.94 ± 0.04, 0.93 ± 0.08, 0.95 ± 0.05, 0.94 ± 0.04, and 0.97, respectively, in a five-fold cross-validation study with pulmonary sarcoidosis (n = 126) and LCa (n = 93) cases. A detailed ablation study showing the impact of CNN + ViT compared to CNN or ViT alone, and CT + radiomics input, compared to CT or radiomics alone, is also presented in this work. Overall, the AI model developed in this work offers promising potential for triaging the pulmonary sarcoidosis patients for timely diagnosis and treatment from chest CT.

An Artificial Intelligence Platform for the Radiologic Diagnosis of Pulmonary Sarcoidosis: An Initial Pilot Study of Chest Computed Tomography Analysis to Distinguish Pulmonary Sarcoidosis from a Negative Lung Cancer Screening Scan.

PURPOSE: To determine the reliability of an artificial intelligence, deep learning (AI/DL)-based method of chest computer tomography (CT) scan analysis to distinguish pulmonary sarcoidosis from negative lung cancer screening chest CT scans (Lung Imaging Reporting and Data System score 1, Lung-RADS score 1). METHODS: Chest CT scans of pulmonary sarcoidosis were evaluated by a clinician experienced with sarcoidosis and a chest radiologist for clinical and radiologic evidence of sarcoidosis and exclusion of alternative or concomitant pulmonary diseases. The AI/DL based method used an ensemble network architecture combining Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). The method was applied to 126 pulmonary sarcoidosis and 96 Lung-RADS score 1 CT scans. The analytic approach of training and validation of the AI/DL method used a fivefold cross-validation technique, where 4/5th of the available data set was used to train a diagnostic model and tested on the remaining 1/5th of the data set, and repeated 4 more times with non-overlapping validation/test data. The probability values were used to generate Receiver Operating Characteristic (ROC) curves to assess the model's discriminatory power. RESULTS: The sensitivity, specificity, positive and negative predictive value of the AI/DL method for the 5 folds of the training/validation sets and the entire set of CT scans were all over 94% to distinguish pulmonary sarcoidosis from LUNG-RADS score 1 chest CT scans. The area under the curve for the corresponding ROC curves were all over 97%. CONCLUSION: This AL/DL model shows promise to distinguish sarcoidosis from alternative pulmonary conditions using minimal radiologic data.

Revealing tumor microstructure with oscillating diffusion encoding MRI in pre-surgical and post-treatment glioma patients.

PURPOSE: We hypothesized that the time-dependent diffusivity at short diffusion times, as measured by oscillating gradient spin echo (OGSE) diffusion MRI, can characterize tissue microstructures in glioma patients. THEORY AND METHODS: Five adult patients with known diffuse glioma, including two pre-surgical and three with new enhancing lesions after treatment for high-grade glioma, were scanned in an ultra-high-performance gradient 3.0T MRI system. OGSE diffusion MRI at 30-100 Hz and pulsed gradient spin echo diffusion imaging (approximated as 0 Hz) were obtained. The ADC and trace-diffusion-weighted image at each acquired frequency were calculated, that is, ADC (f) and TraceDWI (f). RESULTS: In pre-surgical patients, biopsy-confirmed solid enhancing tumor in a high-grade glioblastoma showed higher ADC ( f ) ADC ( 0 Hz ) $$ \frac{\mathrm{ADC}\ (f)}{\mathrm{ADC}\ \left(0\ \mathrm{Hz}\right)} $$ and lower TraceDWI ( f ) TraceDWI ( 0 Hz ) $$ \frac{\mathrm{TraceDWI}\ (f)}{\mathrm{TraceDWI}\ \left(0\ \mathrm{Hz}\right)} $$ , compared to that at same OGSE frequency in a low-grade astrocytoma. In post-treatment patients, the enhancing lesions of two patients who were diagnosed with tumor progression contained more voxels with high ADC ( f ) ADC ( 0 Hz ) $$ \frac{\mathrm{ADC}\ (f)}{\mathrm{ADC}\ \left(0\ \mathrm{Hz}\right)} $$ and low TraceDWI ( f ) TraceDWI ( 0 Hz ) $$ \frac{\mathrm{TraceDWI}\left(\mathrm{f}\right)}{\mathrm{TraceDWI}\left(0\ \mathrm{Hz}\right)} $$ , compared to the enhancing lesions of a patient who was diagnosed with treatment effect. Non-enhancing T2 signal abnormality lesions in both the pre-surgical high-grade glioblastoma and post-treatment tumor progressions showed regions with high ADC ( f ) ADC ( 0 Hz ) $$ \frac{\mathrm{ADC}\ (f)}{\mathrm{ADC}\ \left(0\ \mathrm{Hz}\right)} $$ and low TraceDWI ( f ) TraceDWI ( 0 Hz ) $$ \frac{\mathrm{TraceDWI}\ \left(\mathrm{f}\right)}{\mathrm{TraceDWI}\ \left(0\ \mathrm{Hz}\right)} $$ , consistent with infiltrative tumor. The solid tumor of the glioblastoma, the enhancing lesions of post-treatment tumor progressions, and the suspected infiltrative tumors showed high diffusion time-dependency from 30 to 100 Hz, consistent with high intra-tumoral volume fraction (cellular density). CONCLUSION: Different characteristics of OGSE-based time-dependent diffusivity can reveal heterogenous tissue microstructures that indicate cellular density in glioma patients.

High-Speed On-Site Deep Learning-Based FFR-CT Algorithm: Evaluation Using Invasive Angiography as the Reference Standard.

BACKGROUND. Estimation of fractional flow reserve from coronary CTA (FFR-CT) is an established method of assessing the hemodynamic significance of coronary lesions. However, clinical implementation has progressed slowly, partly because of off-site data transfer with long turnaround times for results. OBJECTIVE. The purpose of this study was to evaluate the diagnostic performance of FFR-CT computed on-site with a high-speed deep learning-based algorithm with invasive hemodynamic indexes as the reference standard. METHODS. This retrospective study included 59 patients (46 men, 13 women; mean age, 66.5 ± 10.2 years) who underwent coronary CTA (including calcium scoring) followed within 90 days by invasive angiography with invasive fractional flow reserve (FFR) and/or instantaneous wave-free ratio measurements from December 2014 to October 2021. Coronary artery lesions were considered to have hemodynamically significant stenosis in the presence of invasive FFR of 0.80 or less and/or instantaneous wave-free ratio of 0.89 or less. A single cardiologist evaluated the CTA images using an on-site deep learning-based semiautomated algorithm entailing a 3D computational flow dynamics model to determine FFR-CT for coronary artery lesions detected with invasive angiography. Time for FFR-CT analysis was recorded. FFR-CT analysis was repeated by the same cardiologist in 26 randomly selected examinations and by a different cardiologist in 45 randomly selected examinations. Diagnostic performance and agreement were assessed. RESULTS. A total of 74 lesions were identified with invasive angiography. FFR-CT and invasive FFR had strong correlation (r = 0.81) and, in Bland-Altman analysis, bias of 0.01 and 95% limits of agreement of -0.13 to 0.15. FFR-CT had AUC for hemodynamically significant stenosis of 0.975. At a cutoff of 0.80 or less, FFR-CT had 95.9% accuracy, 93.5% sensitivity, and 97.7% specificity. In 39 lesions with severe calcifications (≥ 400 Agatston units), FFR-CT had AUC of 0.991 and at a cutoff of 0.80, 94.7% sensitivity, 95.0% specificity, and 94.9% accuracy. Mean analysis time per patient was 7 minutes 54 seconds. Intraobserver agreement (intraclass correlation coefficient, 0.85; bias, -0.01; 95% limits of agreement, -0.12 and 0.10) and interobserver agreement (intraclass correlation coefficient, 0.94; bias, -0.01; 95% limits of agreement, -0.08 and 0.07) were good to excellent. CONCLUSION. A high-speed on-site deep learning-based FFR-CT algorithm had excellent diagnostic performance for hemodynamically significant stenosis with high reproducibility. CLINICAL IMPACT. The algorithm should facilitate implementation of FFR-CT technology into routine clinical practice.

Predicting Breast Cancer Events in Ductal Carcinoma In Situ (DCIS) Using Generative Adversarial Network Augmented Deep Learning Model.

Standard clinicopathological parameters (age, growth pattern, tumor size, margin status, and grade) have been shown to have limited value in predicting recurrence in ductal carcinoma in situ (DCIS) patients. Early and accurate recurrence prediction would facilitate a more aggressive treatment policy for high-risk patients (mastectomy or adjuvant radiation therapy), and simultaneously reduce over-treatment of low-risk patients. Generative adversarial networks (GAN) are a class of DL models in which two adversarial neural networks, generator and discriminator, compete with each other to generate high quality images. In this work, we have developed a deep learning (DL) classification network that predicts breast cancer events (BCEs) in DCIS patients using hematoxylin and eosin (H & E) images. The DL classification model was trained on 67 patients using image patches from the actual DCIS cores and GAN generated image patches to predict breast cancer events (BCEs). The hold-out validation dataset (n = 66) had an AUC of 0.82. Bayesian analysis further confirmed the independence of the model from classical clinicopathological parameters. DL models of H & E images may be used as a risk stratification strategy for DCIS patients to personalize therapy.

A hybrid deformable registration method to generate motion-compensated 3D virtual MRI for fusion with interventional real-time 3D ultrasound.

PURPOSE: Ultrasound is often the preferred modality for image-guided therapy or treatment in organs such as liver due to real-time imaging capabilities. However, the reduced conspicuity of tumors in ultrasound images adversely impacts the precision and accuracy of treatment delivery. This problem is compounded by deformable motion due to breathing and other physiological activity. This creates the need for a fusion method to align interventional US with pre-interventional modalities that provide superior soft-tissue contrast (e.g., MRI) to accurately target a structure-of-interest and compensate for liver motion. METHOD: In this work, we developed a hybrid deformable fusion method to align 3D pre-interventional MRI and 3D interventional US volumes to target the structures-of-interest in liver accurately in real-time. The deformable multimodal fusion method involved an offline alignment of a pre-intervention MRI with a pre-intervention US volume using a traditional registration method, followed by real-time prediction of deformation using a trained deep-learning model between interventional US volumes across different respiratory states. This framework enables motion-compensated MRI-US image fusion in real-time for image-guided treatment. RESULTS: The proposed hybrid deformable registration method was evaluated on three healthy volunteers across the pre-intervention MRI and 20 US volume pairs in the free-breathing respiratory cycle. The mean Euclidean landmark distance of three homologous targets in all three volunteers was less than 3 mm for percutaneous liver procedures. CONCLUSIONS: Preliminary results show that clinically acceptable registration accuracies for near real-time, deformable MRI-US fusion can be achieved by our proposed hybrid approach. The proposed combination of traditional and deep-learning deformable registration techniques is thus a promising approach for motion-compensated MRI-US fusion to improve targeting in image-guided liver interventions.

Automatic Hemorrhage Detection From Color Doppler Ultrasound Using a Generative Adversarial Network (GAN)-Based Anomaly Detection Method.

Hemorrhage control has been identified as a priority focus area both for civilian and military populations in the United States because exsanguination is the most common cause of preventable death in hemorrhagic injury. Non-compressible torso hemorrhage (NCTH) has high mortality rate and there are currently no broadly available therapies for NCTH outside of a surgical room environment. Novel therapies, which include High Intensity Focused Ultrasound (HIFU) have emerged as promising methods for hemorrhage control as they can non-invasively cauterize bleeding tissue deep within the body without injuring uninvolved regions. A major challenge in the application of HIFU with color Doppler US guidance is the interpretation and optimization of the blood flow images in real-time to identify the hemorrhagic focus. Today, this task requires an expert sonographer, limiting the utility of this therapy in non-clinical environments. In this work, we investigated the feasibility of an automated hemorrhage detection method using a Generative Adversarial Network (GAN) for anomaly detection that learns a manifold of normal blood flow variability and subsequently identifies anomalous flow patterns that fall outside the learned manifold. As an initial feasibility study, we collected ultrasound color Doppler images of femoral arteries in an animal model of vascular injury (N = 11 pigs). Velocity information of the blood flow were extracted from the color Doppler images that were used for training and testing the anomaly detection network. Normotensive images from 8 pigs were used for training, and testing was performed on normotensive, immediately after injury, 10 minutes post-injury and 30 minutes post-injury images from 3 other pigs. The residual images or the reconstructed error maps show promise in detecting hemorrhages with an AUC of 0.90, 0.87, 0.62 immediately, 10 minutes post-injury and 30 minutes post-injury respectively with an overall AUC of 0.83.

First-in-human imaging using a MR-compatible e4D ultrasound probe for motion management of radiotherapy.

PURPOSE: Respiration-induced tumor or organ positional changes can impact the accuracy of external beam radiotherapy. Motion management strategies are used to account for these changes during treatment. The authors report on the development, testing, and first-in-human evaluation of an electronic 4D (e4D) MR-compatible ultrasound probe that was designed for hands-free operation in a MR and linear accelerator (LINAC) environment. METHODS: Ultrasound components were evaluated for MR compatibility. Electromagnetic interference (EMI) shielding was used to enclose the entire probe and a factory-fabricated cable shielded with copper braids was integrated into the probe. A series of simultaneous ultrasound and MR scans were acquired and analyzed in five healthy volunteers. RESULTS: The ultrasound probe led to minor susceptibility artifacts in the MR images immediately proximal to the ultrasound probe at a depth of <10 mm. Ultrasound and MR-based motion traces that were derived by tracking the salient motion of endogenous target structures in the superior-inferior (SI) direction demonstrated good concordance (Pearson correlation coefficients of 0.95-0.98) between the ultrasound and MRI datasets. CONCLUSION: We have demonstrated that our hands-free, e4D probe can acquire ultrasound images during a MR acquisition at frame rates of approximately 4 frames per second (fps) without impacting either the MR or ultrasound image quality. This use of this technology for interventional procedures (e.g. biopsies and drug delivery) and motion compensation during imaging are also being explored.

Oscillating diffusion-encoding with a high gradient-amplitude and high slew-rate head-only gradient for human brain imaging.

PURPOSE: We investigate the importance of high gradient-amplitude and high slew-rate on oscillating gradient spin echo (OGSE) diffusion imaging for human brain imaging and evaluate human brain imaging with OGSE on the MAGNUS head-gradient insert (200 mT/m amplitude and 500 T/m/s slew rate). METHODS: Simulations with cosine-modulated and trapezoidal-cosine OGSE at various gradient amplitudes and slew rates were performed. Six healthy subjects were imaged with the MAGNUS gradient at 3T with OGSE at frequencies up to 100 Hz and b = 450 s/mm2 . Comparisons were made against standard pulsed gradient spin echo (PGSE) diffusion in vivo and in an isotropic diffusion phantom. RESULTS: Simulations show that to achieve high frequency and b-value simultaneously for OGSE, high gradient amplitude, high slew rates, and high peripheral nerve stimulation limits are required. A strong linear trend for increased diffusivity (mean: 8-19%, radial: 9-27%, parallel: 8-15%) was observed in normal white matter with OGSE (20 Hz to 100 Hz) as compared to PGSE. Linear fitting to frequency provided excellent correlation, and using a short-range disorder model provided radial long-term diffusivities of D∞,MD = 911 ± 72 µm2 /s, D∞,PD = 1519 ± 164 µm2 /s, and D∞,RD = 640 ± 111 µm2 /s and correlation lengths of lc,MD = 0.802 ± 0.156 µm, lc,PD = 0.837 ± 0.172 µm, and lc,RD = 0.780 ± 0.174 µm. Diffusivity changes with OGSE frequency were negligible in the phantom, as expected. CONCLUSION: The high gradient amplitude, high slew rate, and high peripheral nerve stimulation thresholds of the MAGNUS head-gradient enables OGSE acquisition for in vivo human brain imaging.

Mass Effect Deformation Heterogeneity (MEDH) on Gadolinium-contrast T1-weighted MRI is associated with decreased survival in patients with right cerebral hemisphere Glioblastoma: A feasibility study.

Subtle tissue deformations caused by mass-effect in Glioblastoma (GBM) are often not visually evident, and may cause neurological deficits, impacting survival. Radiomic features provide sub-visual quantitative measures to uncover disease characteristics. We present a new radiomic feature to capture mass effect-induced deformations in the brain on Gadolinium-contrast (Gd-C) T1w-MRI, and their impact on survival. Our rationale is that larger variations in deformation within functionally eloquent areas of the contralateral hemisphere are likely related to decreased survival. Displacements in the cortical and subcortical structures were measured by aligning the Gd-C T1w-MRI to a healthy atlas. The variance of deformation magnitudes was measured and defined as Mass Effect Deformation Heterogeneity (MEDH) within the brain structures. MEDH values were then correlated with overall-survival of 89 subjects on the discovery cohort, with tumors on the right (n = 41) and left (n = 48) cerebral hemispheres, and evaluated on a hold-out cohort (n = 49 subjects). On both cohorts, decreased survival time was found to be associated with increased MEDH in areas of language comprehension, social cognition, visual perception, emotion, somato-sensory, cognitive and motor-control functions, particularly in the memory areas in the left-hemisphere. Our results suggest that higher MEDH in functionally eloquent areas of the left-hemisphere due to GBM in the right-hemisphere may be associated with poor-survival.

Prostate shapes on pre-treatment MRI between prostate cancer patients who do and do not undergo biochemical recurrence are different: Preliminary Findings.

Early identification of PCa patients at risk for biochemical recurrence (BCR) post-therapy will potentially complement definitive therapy with either neo- or adjuvant therapy to improve prognosis. BCR post definitive therapy is often associated with disease progression that might cause a bulge in the prostate gland. In this work we explored if an atlas-based comparison approach reveals shape differences in the prostate capsule as observed on pre-treatment T2-weighted MRI between prostate cancer patients who do (BCR +) and do not (BCR -) have BCR following definitive therapy. A single center IRB approved study included 874 patients. Complete image datasets, clinically localized PCa, availability of Gleason score, data available for post-treatment PSA and follow-up for at least 3 years in patients without BCR were the inclusion criteria to select 77 patients out of the 874 patients. Further controlling for Gleason score, stage, age and to maintain equal number of cases for the BCR + and BCR - categories, the total number of cases was reduced to 50. Manually segmented prostate capsules were aligned to a BCR - template for statistical comparison between the BCR + and BCR - groups. Statistically significant shape difference between the two groups was observed towards the lateral and the posterior sides of prostate.

Regression and statistical shape model based substitute CT generation for MRI alone external beam radiation therapy from standard clinical MRI sequences.

In MR only radiation therapy planning, generation of the tissue specific HU map directly from the MRI would eliminate the need of CT image acquisition and may improve radiation therapy planning. The aim of this work is to generate and validate substitute CT (sCT) scans generated from standard T2 weighted MR pelvic scans in prostate radiation therapy dose planning. A Siemens Skyra 3T MRI scanner with laser bridge, flat couch and pelvic coil mounts was used to scan 39 patients scheduled for external beam radiation therapy for localized prostate cancer. For sCT generation a whole pelvis MRI (1.6 mm 3D isotropic T2w SPACE sequence) was acquired. Patients received a routine planning CT scan. Co-registered whole pelvis CT and T2w MRI pairs were used as training images. Advanced tissue specific non-linear regression models to predict HU for the fat, muscle, bladder and air were created from co-registered CT-MRI image pairs. On a test case T2w MRI, the bones and bladder were automatically segmented using a novel statistical shape and appearance model, while other soft tissues were separated using an Expectation-Maximization based clustering model. The CT bone in the training database that was most 'similar' to the segmented bone was then transformed with deformable registration to create the sCT component of the test case T2w MRI bone tissue. Predictions for the bone, air and soft tissue from the separate regression models were successively combined to generate a whole pelvis sCT. The change in monitor units between the sCT-based plans relative to the gold standard CT plan for the same IMRT dose plan was found to be [Formula: see text] (mean ± standard deviation) for 39 patients. The 3D Gamma pass rate was [Formula: see text] (2 mm/2%). The novel hybrid model is computationally efficient, generating an sCT in 20 min from standard T2w images for prostate cancer radiation therapy dose planning and DRR generation.

Multi-atlas and unsupervised learning approach to perirectal space segmentation in CT images.

Perirectal space segmentation in computed tomography images aids in quantifying radiation dose received by healthy tissues and toxicity during the course of radiation therapy treatment of the prostate. Radiation dose normalised by tissue volume facilitates predicting outcomes or possible harmful side effects of radiation therapy treatment. Manual segmentation of the perirectal space is time consuming and challenging in the presence of inter-patient anatomical variability and may suffer from inter- and intra-observer variabilities. However automatic or semi-automatic segmentation of the perirectal space in CT images is a challenging task due to inter patient anatomical variability, contrast variability and imaging artifacts. In the model presented here, a volume of interest is obtained in a multi-atlas based segmentation approach. Un-supervised learning in the volume of interest with a Gaussian-mixture-modeling based clustering approach is adopted to achieve a soft segmentation of the perirectal space. Probabilities from soft clustering are further refined by rigid registration of the multi-atlas mask in a probabilistic domain. A maximum a posteriori approach is adopted to achieve a binary segmentation from the refined probabilities. A mean volume similarity value of 97% and a mean surface difference of 3.06 ± 0.51 mm is achieved in a leave-one-patient-out validation framework with a subset of a clinical trial dataset. Qualitative results show a good approximation of the perirectal space volume compared to the ground truth.

Neuroanatomical correlates of childhood apraxia of speech: A connectomic approach.

Childhood apraxia of speech (CAS) is a paediatric speech sound disorder in which precision and consistency of speech movements are impaired. Most children with idiopathic CAS have normal structural brain MRI. We hypothesize that children with CAS have altered structural connectivity in speech/language networks compared to controls and that these altered connections are related to functional speech/language measures. Whole brain probabilistic tractography, using constrained spherical deconvolution, was performed for connectome generation in 17 children with CAS and 10 age-matched controls. Fractional anisotropy (FA) was used as a measure of connectivity and the connections with altered FA between CAS and controls were identified. Further, the relationship between altered FA and speech/language scores was determined. Three intra-hemispheric/interhemispheric subnetworks showed reduction of FA in CAS compared to controls, including left inferior (opercular part) and superior (dorsolateral, medial and orbital part) frontal gyrus, left superior and middle temporal gyrus and left post-central gyrus (subnetwork 1); right supplementary motor area, left middle and inferior (orbital part) frontal gyrus, left precuneus and cuneus, right superior occipital gyrus and right cerebellum (subnetwork 2); right angular gyrus, right superior temporal gyrus and right inferior occipital gyrus (subnetwork 3). Reduced FA of some connections correlated with diadochokinesis, oromotor skills, expressive grammar and poor lexical production in CAS. These findings provide evidence of structural connectivity anomalies in children with CAS across specific brain regions involved in speech/language function. We propose altered connectivity as a possible epiphenomenon of complex pathogenic mechanisms in CAS which need further investigation.

MRI-alone radiation therapy planning for prostate cancer: Automatic fiducial marker detection.

PURPOSE: The feasibility of radiation therapy treatment planning using substitute computed tomography (sCT) generated from magnetic resonance images (MRIs) has been demonstrated by a number of research groups. One challenge with an MRI-alone workflow is the accurate identification of intraprostatic gold fiducial markers, which are frequently used for prostate localization prior to each dose delivery fraction. This paper investigates a template-matching approach for the detection of these seeds in MRI. METHODS: Two different gradient echo T1 and T2* weighted MRI sequences were acquired from fifteen prostate cancer patients and evaluated for seed detection. For training, seed templates from manual contours were selected in a spectral clustering manifold learning framework. This aids in clustering "similar" gold fiducial markers together. The marker with the minimum distance to a cluster centroid was selected as the representative template of that cluster during training. During testing, Gaussian mixture modeling followed by a Markovian model was used in automatic detection of the probable candidates. The probable candidates were rigidly registered to the templates identified from spectral clustering, and a similarity metric is computed for ranking and detection. RESULTS: A fiducial detection accuracy of 95% was obtained compared to manual observations. Expert radiation therapist observers were able to correctly identify all three implanted seeds on 11 of the 15 scans (the proposed method correctly identified all seeds on 10 of the 15). CONCLUSIONS: An novel automatic framework for gold fiducial marker detection in MRI is proposed and evaluated with detection accuracies comparable to manual detection. When radiation therapists are unable to determine the seed location in MRI, they refer back to the planning CT (only available in the existing clinical framework); similarly, an automatic quality control is built into the automatic software to ensure that all gold seeds are either correctly detected or a warning is raised for further manual intervention.

Statistical machine learning to identify traumatic brain injury (TBI) from structural disconnections of white matter networks.

Identifying diffuse axonal injury (DAI) in patients with traumatic brain injury (TBI) presenting with normal appearing radiological MRI presents a significant challenge. Neuroimaging methods such as diffusion MRI and probabilistic tractography, which probe the connectivity of neural networks, show significant promise. We present a machine learning approach to classify TBI participants primarily with mild traumatic brain injury (mTBI) based on altered structural connectivity patterns derived through the network based statistical analysis of structural connectomes generated from TBI and age-matched control groups. In this approach, higher order diffusion models were used to map white matter connections between 116 cortical and subcortical regions. Tracts between these regions were generated using probabilistic tracking and mean fractional anisotropy (FA) measures along these connections were encoded in the connectivity matrices. Network-based statistical analysis of the connectivity matrices was performed to identify the network differences between a representative subset of the two groups. The affected network connections provided the feature vectors for principal component analysis and subsequent classification by random forest. The validity of the approach was tested using data acquired from a total of 179 TBI patients and 146 controls participants. The analysis revealed altered connectivity within a number of intra- and inter-hemispheric white matter pathways associated with DAI, in consensus with existing literature. A mean classification accuracy of 68.16%±1.81% and mean sensitivity of 80.0%±2.36% were achieved in correctly classifying the TBI patients evaluated on the subset of the participants that was not used for the statistical analysis, in a 10-fold cross-validation framework. These results highlight the potential for statistical machine learning approaches applied to structural connectomes to identify patients with diffusive axonal injury.

An Improved Patient-Specific Mortality Risk Prediction in ICU in a Random Forest Classification Framework.

Dynamic and automatic patient specific prediction of the risk associated with ICU mortality may facilitate timely and appropriate intervention of health professionals in hospitals. In this work, patient information and time series measurements of vital signs and laboratory results from the first 48 hours of ICU stays of 4000 adult patients from a publicly available dataset are used to design and validate a mortality prediction system. An ensemble of decision trees are used to simultaneously predict and associate a risk score against each patient in a k-fold validation framework. Risk assessment prediction accuracy of 87% is achieved with our model and the results show significant improvement over a baseline algorithm of SAPS-I that is commonly used for mortality prediction in ICU. The performance of our model is further compared to other state-of-the-art algorithms evaluated on the same dataset.

Human Activity Recognition from Smart-Phone Sensor Data using a Multi-Class Ensemble Learning in Home Monitoring.

Home monitoring of chronically ill or elderly patient can reduce frequent hospitalisations and hence provide improved quality of care at a reduced cost to the community, therefore reducing the burden on the healthcare system. Activity recognition of such patients is of high importance in such a design. In this work, a system for automatic human physical activity recognition from smart-phone inertial sensors data is proposed. An ensemble of decision trees framework is adopted to train and predict the multi-class human activity system. A comparison of our proposed method with a multi-class traditional support vector machine shows significant improvement in activity recognition accuracies.

Lesion segmentation from multimodal MRI using random forest following ischemic stroke.

Understanding structure-function relationships in the brain after stroke is reliant not only on the accurate anatomical delineation of the focal ischemic lesion, but also on previous infarcts, remote changes and the presence of white matter hyperintensities. The robust definition of primary stroke boundaries and secondary brain lesions will have significant impact on investigation of brain-behavior relationships and lesion volume correlations with clinical measures after stroke. Here we present an automated approach to identify chronic ischemic infarcts in addition to other white matter pathologies, that may be used to aid the development of post-stroke management strategies. Our approach uses Bayesian-Markov Random Field (MRF) classification to segment probable lesion volumes present on fluid attenuated inversion recovery (FLAIR) MRI. Thereafter, a random forest classification of the information from multimodal (T1-weighted, T2-weighted, FLAIR, and apparent diffusion coefficient (ADC)) MRI images and other context-aware features (within the probable lesion areas) was used to extract areas with high likelihood of being classified as lesions. The final segmentation of the lesion was obtained by thresholding the random forest probabilistic maps. The accuracy of the automated lesion delineation method was assessed in a total of 36 patients (24 male, 12 female, mean age: 64.57±14.23yrs) at 3months after stroke onset and compared with manually segmented lesion volumes by an expert. Accuracy assessment of the automated lesion identification method was performed using the commonly used evaluation metrics. The mean sensitivity of segmentation was measured to be 0.53±0.13 with a mean positive predictive value of 0.75±0.18. The mean lesion volume difference was observed to be 32.32%±21.643% with a high Pearson's correlation of r=0.76 (p<0.0001). The lesion overlap accuracy was measured in terms of Dice similarity coefficient with a mean of 0.60±0.12, while the contour accuracy was observed with a mean surface distance of 3.06mm±3.17mm. The results signify that our method was successful in identifying most of the lesion areas in FLAIR with a low false positive rate.

Evaluation of prostate segmentation algorithms for MRI: the PROMISE12 challenge.

Prostate MRI image segmentation has been an area of intense research due to the increased use of MRI as a modality for the clinical workup of prostate cancer. Segmentation is useful for various tasks, e.g. to accurately localize prostate boundaries for radiotherapy or to initialize multi-modal registration algorithms. In the past, it has been difficult for research groups to evaluate prostate segmentation algorithms on multi-center, multi-vendor and multi-protocol data. Especially because we are dealing with MR images, image appearance, resolution and the presence of artifacts are affected by differences in scanners and/or protocols, which in turn can have a large influence on algorithm accuracy. The Prostate MR Image Segmentation (PROMISE12) challenge was setup to allow a fair and meaningful comparison of segmentation methods on the basis of performance and robustness. In this work we will discuss the initial results of the online PROMISE12 challenge, and the results obtained in the live challenge workshop hosted by the MICCAI2012 conference. In the challenge, 100 prostate MR cases from 4 different centers were included, with differences in scanner manufacturer, field strength and protocol. A total of 11 teams from academic research groups and industry participated. Algorithms showed a wide variety in methods and implementation, including active appearance models, atlas registration and level sets. Evaluation was performed using boundary and volume based metrics which were combined into a single score relating the metrics to human expert performance. The winners of the challenge where the algorithms by teams Imorphics and ScrAutoProstate, with scores of 85.72 and 84.29 overall. Both algorithms where significantly better than all other algorithms in the challenge (p<0.05) and had an efficient implementation with a run time of 8min and 3s per case respectively. Overall, active appearance model based approaches seemed to outperform other approaches like multi-atlas registration, both on accuracy and computation time. Although average algorithm performance was good to excellent and the Imorphics algorithm outperformed the second observer on average, we showed that algorithm combination might lead to further improvement, indicating that optimal performance for prostate segmentation is not yet obtained. All results are available online at http://promise12.grand-challenge.org/.