Machine learning-based analysis of 68Ga-PSMA-11 PET/CT images for estimation of prostate tumor grade.

Early diagnosis of prostate cancer, the most common malignancy in men, can improve patient outcomes. Since the tissue sampling procedures are invasive and sometimes inconclusive, an alternative image-based method can prevent possible complications and facilitate treatment management. We aim to propose a machine-learning model for tumor grade estimation based on 68 Ga-PSMA-11 PET/CT images in prostate cancer patients. This study included 90 eligible participants out of 244 biopsy-proven prostate cancer patients who underwent staging 68Ga-PSMA-11 PET/CT imaging. The patients were divided into high and low-intermediate groups based on their Gleason scores. The PET-only images were manually segmented, both lesion-based and whole prostate, by two experienced nuclear medicine physicians. Four feature selection algorithms and five classifiers were applied to Combat-harmonized and non-harmonized datasets. To evaluate the model's generalizability across different institutions, we performed leave-one-center-out cross-validation (LOOCV). The metrics derived from the receiver operating characteristic curve were used to assess model performance. In the whole prostate segmentation, combining the ANOVA algorithm as the feature selector with Random Forest (RF) and Extra Trees (ET) classifiers resulted in the highest performance among the models, with an AUC of 0.78 and 083, respectively. In the lesion-based segmentation, the highest AUC was achieved by MRMR feature selector + Linear Discriminant Analysis (LDA) and Logistic Regression (LR) classifiers (0.76 and 0.79, respectively). The LOOCV results revealed that both the RF_ANOVA and ET_ANOVA models showed high levels of accuracy and generalizability across different centers, with an average AUC value of 0.87 for the ET_ANOVA combination. Machine learning-based analysis of radiomics features extracted from 68Ga-PSMA-11 PET/CT scans can accurately classify prostate tumors into low-risk and intermediate- to high-risk groups.

Comparison of the ocular ultrasonic and optical biometry devices in the different quality measurements.

PURPOSE: To compare the reliability and agreement of axial length (AL), anterior chamber depth (ACD), and lens thickness (LT) measurements obtained with optical biometry based on swept-source optical coherence tomography (IOLMaster 700; Carl Zeiss, Germany) and an ultrasound biometry device (Nidek; US-4000 Echoscan, Japan) in different qualities of AL measurement. METHODS: A total of 239 consecutive eyes of 239 cataract surgery candidates with a mean age of 56 ± 14 years were included. The quality measurements were grouped according to the quartiles of SD of the measured AL by IOLMaster 700. The first and fourth quartile's SD are defined as high and low-quality measurement, respectively, and the second and third quartiles' SD is defined as moderate-quality. RESULTS: The reliability of AL and ACD between the two devices in all patients and in different quality measurement groups was excellent with highly statistically significant (AL: all ICC=0.999 and P<0.001, ACD: all ICC>0.920 and P<0.001). AL and ACD in all quality measurements showed a very strong correlation between devices with highly statistically significant. However, there was poor (ICC=0.305), moderate (ICC=0.742), and good (ICC=0.843) reliability in measuring LT in low-, moderate-, and high-quality measurements, respectively. LT showed a very strong correlation (r = 0.854) with highly statistically significant (P<0.001) between devices only in patients with high-quality measurements. CONCLUSIONS: AL and ACD of the IOLMaster700 had outstanding agreements with the US-4000 ultrasound in different quality measurements of AL and can be used interchangeably. But LT should be used interchangeably cautiously only in the high-quality measurements group.

An automatic segmentation framework for computer-assisted renal scintigraphy procedure.

One of the techniques for achieving unique and reliable information in medicine is renal scintigraphy. A key step for quantitative renal scintigraphy is segmentation of the kidneys. Here, an automatic segmentation framework was proposed for computer-aided renal scintigraphy procedures. To extract kidney boundary in dynamic renal scintigraphic images, a multi-step approach was proposed. This technique is featured with key steps, namely, localization and segmentation. At first, the ROI of each kidney was estimated using Otsu's thresholding, anatomical constraint, and integral projection, which is done in an automatic process. Afterwards, the ROI obtained for the kidneys was used as the initial contours to create the final counter of kidneys using geometric active contours. At this step and for the segmentation, an improved variational level set was utilized through Mumford-Shah formulation. Using e.cam gamma camera system (SIEMENS), 30 data sets were used to assess the proposed method. By comparing the manually outlined borders, the performance of the proposed method was shown. Different measures were used to examine the performance. It was found that the proposed segmentation method managed to extract the kidney boundary in renal scintigraphic images. The proposed technique achieved a sensitivity of 95.15% and a specificity of 95.33%. In addition, the section under the curve in the ROC analysis was equal to 0.974. The proposed technique successfully segmented the renal contour in dynamic renal scintigraphy. Using all the data sets, a correct segmentation of the kidney was performed. In addition, the technique was successful with noisy and low-resolution images and challenging cases with close interfering activities such as liver and spleen activities.

Grading of gliomas using transfer learning on MRI images.

OBJECTIVE: Despite the critical role of Magnetic Resonance Imaging (MRI) in the diagnosis of brain tumours, there are still many pitfalls in the exact grading of them, in particular, gliomas. In this regard, it was aimed to examine the potential of Transfer Learning (TL) and Machine Learning (ML) algorithms in the accurate grading of gliomas on MRI images. MATERIALS AND METHODS: Dataset has included four types of axial MRI images of glioma brain tumours with grades I-IV: T1-weighted, T2-weighted, FLAIR, and T1-weighted Contrast-Enhanced (T1-CE). Images were resized, normalized, and randomly split into training, validation, and test sets. ImageNet pre-trained Convolutional Neural Networks (CNNs) were utilized for feature extraction and classification, using Adam and SGD optimizers. Logistic Regression (LR) and Support Vector Machine (SVM) methods were also implemented for classification instead of Fully Connected (FC) layers taking advantage of features extracted by each CNN. RESULTS: Evaluation metrics were computed to find the model with the best performance, and the highest overall accuracy of 99.38% was achieved for the model containing an SVM classifier and features extracted by pre-trained VGG-16. DISCUSSION: It was demonstrated that developing Computer-aided Diagnosis (CAD) systems using pre-trained CNNs and classification algorithms is a functional approach to automatically specify the grade of glioma brain tumours in MRI images. Using these models is an excellent alternative to invasive methods and helps doctors diagnose more accurately before treatment.

Radiomics-based machine-learning method to diagnose prostate cancer using mp-MRI: a comparison between conventional and fused models.

OBJECTIVES: Multiparametric MRI (mp-MRI) has been significantly used for detection, localization and staging of Prostate cancer (PCa). However, all the assessment suffers from poor reproducibility among the readers. The aim of this study was to evaluate radiomics models to diagnose PCa using high-resolution T2-weighted (T2-W) and dynamic contrast-enhanced (DCE) MRI. MATERIALS AND METHODS: Thirty two patients who had high prostate specific antigen level were recruited. The prostate biopsies considered as the reference to differentiate between 66 benign and 36 malignant prostate lesions. 181 features were extracted from each modality. K-nearest neighbors, artificial neural network, decision tree, and linear discriminant analysis were used for machine-learning study. The leave-one-out cross-validation method was used to prevent overfitting and build robust models. RESULTS: Radiomics analysis showed that T2-W images were more effective in PCa detection compare to DCE images. Local binary pattern features and speeded up robust features had the highest ability for prediction in T2-W and DCE images, respectively. The classifier fusion using decision template method showed the highest performance with accuracy, specificity, and sensitivity of 100%. DISCUSSION: The findings of this framework provide researchers on PCa with a promising method for reliable detection of prostate lesions in MR images by fused model.

Organ dose in cardiac dual-energy computed tomography: a Monte Carlo study.

Dual-energy computed tomography (DECT) has appeared as a novel approach with the aim of evaluating artery-related diseases. With the advent of DECT, concerns have been raised about the induction of diseases such as cancer due to high radiation exposure of patients. Therefore, the dose received by patients in DECT should be considered. The parameter most commonly used for patient dosimetry is the effective dose (ED). The purpose of this study is to model and validate a DECT scanner by a developed MCNP Monte Carlo code and to calculate the organ doses, the ED, and the conversion factor (k-factor) used in determining ED in the cardiac imaging protocol. To validate the DECT scanner simulation, a standard dosimetry body phantom was modeled in two radiation modes of single energy CT and DECT. The results of simulated CT dose index (CTDI) were compared with those of ImPACT or measurement data. Then dosimetry phantom was replaced by the male and female ORNL phantoms and the organ doses were calculated. The organ doses were also calculated by ImPACT dose software. In the initial validation stage, the minimum and maximum observed relative differences between results of MNCP simulation and measured were 2.77% and 5.79% for the central CTDI and 1.91% and 5.83% for the averaged peripheral CTDI, respectively. The mean ED of simulation and the ImPACT were 3.23 and 5.55 mSv/100 mAs, and the mean k-factor was 0.016 and 0.032 mSv mGy-1 cm-1 in the male and female phantoms, respectively. The k-factor obtained for males is close to the currently used k-factor, but the k-factor for females is almost twice.

A Monte Carlo Platform for Characterization of X-Ray Radiation Dose in CT Imaging.

BACKGROUND: Computed tomography (CT) is currently known as a versatile imaging tool in the clinic used for almost all types of cancers. The major issue of CT is the health risk, belonging to X-ray radiation exposure. Concerning this, Monte Carlo (MC) simulation is recognized as a key computational technique for estimating and optimizing radiation dose. CT simulation with MCNP/MCNPX MC code has an inherent problem due to the lack of a fan-beam shaped source model. This limitation increases the run time and highly decreases the number of photons passing the body or phantom. Recently, a beta version of MCNP code called MCNP-FBSM (Fan-Beam Source Model) has been developed to pave the simulation way of CT imaging procedure, removing the need of the collimator. This is a new code, which needs to be validated in all aspects. OBJECTIVE: In this work, we aimed to develop and validate an efficient computational platform based on modified MCNP-FBSM for CT dosimetry purposes. MATERIAL AND METHODS: In this experimental study, a setup is carried out to measure CTDI100 in air and standard dosimetry phantoms. The accuracy of the developed MC CT simulator results has been widely benchmarked through comparison with our measured data, UK's National Health Service's reports (known as ImPACT), manufacturer's data, and other published results. RESULTS: The minimum and maximum observed mean differences of our simulation results and other above-mentioned data were the 1.5%, and 9.79%, respectively. CONCLUSION: The developed FBSM MC computational platform is a beneficial tool for CT dosimetry.

Parametric-based feature selection via spherical harmonic coefficients for the left ventricle myocardial infarction screening.

Computer-aided diagnosis (CAD) of heart diseases using machine learning techniques has recently received much attention. In this study, we present a novel parametric-based feature selection method using the three-dimensional spherical harmonic (SHs) shape descriptors of the left ventricle (LV) for intelligent myocardial infarction (MI) classification. The main hypothesis is that the SH coefficients of the parameterized endocardial shapes in MI patients are recognizable and distinguishable from healthy subjects. The SH parameterization, expansion, and registration of the LV endocardial shapes were performed, then parametric-based features were extracted. The proposed method performance was investigated by varying considered phases (i.e., the end-systole (ES) or the end-diastole (ED) frames), the spatial alignment procedures based on three modes (i.e., the center of the apical (CoA), the center of mass (CoM), and the center of the basal (CoB)), and considered orders of SH coefficients. After applying principal component analysis (PCA) on the feature vectors, support vector machine (SVM), K-nearest neighbors (K-NN), and random forest (RF) were trained and tested using the leave-one-out cross-validation (LOOCV). The proposed method validation was performed via a dataset containing healthy and MI subjects selected from the automated cardiac diagnosis challenge (ACDC) database. The promising results show the effectiveness of the proposed classification model. SVM reached the best performance with accuracy, sensitivity, specificity, and F-score of 97.50%, 95.00%, 100.00%, and 97.56%, respectively, using the introduced optimum feature set. This study demonstrates the robustness of combining the SH coefficients and machine learning techniques. We also quantify and notably highlight the contribution of different parameters in the classification and finally introduce an optimal feature set with maximum discriminant strength for the MI classification task. Moreover, the obtained results confirm that the proposed method performs more accurately than conventional point-based methods and also the current start-of-the-art, i.e., clinical measures. We showed our method's generalizability using employing it in dilated cardiomyopathy (DCM) detection and achieving promising results too. Parametric-based feature selection via spherical harmonics coefficients for the left ventricle myocardial infarction screening.

2D Statistical Lung Shape Analysis Using Chest Radiographs: Modelling and Segmentation.

Accurate information of the lung shape analysis and its anatomical variations is very noticeable in medical imaging. The normal variations of the lung shape can be interpreted as a normal lung. In contrast, abnormal variations of the lung shape can be a result of one of the pulmonary diseases. The goal of this study is twofold: (1) represent two lung shape models which are different at the reference points in registration process considering to show their impact on estimating the inter-patient 2D lung shape variations and (2) using the obtained models in lung field segmentation by utilizing active shape model (ASM) technique. The represented models which showed the inter-patient 2D lung shape variations in two different forms are fully compared and evaluated. The results show that the models along with standard principal component analysis (PCA) can be able to explain more than 95% of total variations in all cases using only first 7 principal component (PC) modes for both lungs. Both models are used in ASM-based segmentation technique for lung field segmentation. The segmentation results are evaluated using leave-one-out cross validation technique. According to the experimental results, the proposed method has average dice similarity coefficient of 97.1% and 96.1% for the right and the left lung, respectively. The results show that the proposed segmentation method is more stable and accurate than other model-based techniques to inter-patient lung field segmentation.

Contour-based lung shape analysis in order to tuberculosis detection: modeling and feature description.

Statistical shape analysis of lung is a reliable alternative method for diagnosing pulmonary diseases such as tuberculosis (TB). The 2D contour-based lung shape analysis is investigated and developed using Fourier descriptors (FDs). The proposed 2D lung shape analysis is carried out in threefold: (1) represent the normal and the abnormal (i.e. pulmonary tuberculosis (PTB)) lung shape models using Fourier descriptors modeling (FDM) framework from chest X-ray (CXR) images, (2) estimate and compare the 2D inter-patient lung shape variations for the normal and abnormal lungs by applying principal component analysis (PCA) techniques, and (3) describe the optimal type of contour-based feature vectors to train a classifier in order to detect TB using one publicly available dataset-namely the Montgomery dataset. Since almost all of the previous works in lung shape analysis are content-based analysis, we proposed contour-based lung shape analysis for statistical modeling and feature description of PTB cases. The results show that the proposed approach is able to explain more than 95% of total variations in both of the normal and PTB cases using only 6 and 7 principal component modes for the right and the left lungs, respectively. In case of PTB detection, using 138 lung cases (80 normal and 58 PTB cases), we achieved the accuracy (ACC) and the area under the curve (AUC) of 82.03% and 88.75%, respectively. In comparison with existing state-of-art studies in the same dataset, the proposed approach is a very promising supplement for diagnosis of PTB disease. The method is robust and valuable for application in 2D automatic segmentation, classification, and atlas registration. Moreover, the approach could be used for any kind of pulmonary diseases. Graphical abstract Contour-based lung shape analysis in order to detect tuberculosis: modeling and feature description.

Inter-Patient Modelling of 2D Lung Variations from Chest X-Ray Imaging via Fourier Descriptors.

Detailed knowledge of anatomical lung variation is very important in medical image processing. Normal variations of lung consistent with the maintenance of pulmonary health and abnormal lung variations can be as a result of a pulmonary disease. Inter-patient lung variations can be due to the several factors such as sex, age, height, weight and type of disease. This study tries to show the inter-patient lung variations by using one of the shape-based descriptions techniques which is called Fourier descriptors. Shape-based description is an important approach to construct an object according to its parametric values. A different types of techniques are reported in the literature that aim to represent objects based on their shapes; each of these techniques has its cons and pros. Fourier descriptors, a simple yet powerful technique, has interesting properties such as rotational, scale, and translational invariance and these are powerful features for the recognition of two-dimensional connected shapes. In this paper, we use 380 CXR (Chest X-ray) images as a training set to construct the statistical mean model of lung contour. For modelling, the first step is evaluation of lung contour approximation and characterization to get the good spatial and frequency resolution. In the second step, all of the lung contours registered to show the variation and make a mean shape (i.e. lungs). And the final step is calculating the dispersion (i.e. covariance matrix) and analyzing by principle components. The proposed technique used to create the inter-patient statistical model and provide statistical parameters for application in segmentation, classification, 2D atlas based registration, etc. In this paper, we presented an approach for creating 2D modelling of human lungs from CXR image archives and reported some interesting statistical parameters to analysis the left and the right lung shape.

Magnetic resonance imaging-based target volume delineation in radiation therapy treatment planning for brain tumors using localized region-based active contour.

PURPOSE: To evaluate the clinical application of a robust semiautomatic image segmentation method to determine the brain target volumes in radiation therapy treatment planning. METHODS AND MATERIALS: A local robust region-based algorithm was used on MRI brain images to study the clinical target volume (CTV) of several patients. First, 3 oncologists delineated CTVs of 10 patients manually, and the process time for each patient was calculated. The averages of the oncologists' contours were evaluated and considered as reference contours. Then, to determine the CTV through the semiautomatic method, a fourth oncologist who was blind to all manual contours selected 4-8 points around the edema and defined the initial contour. The time to obtain the final contour was calculated again for each patient. Manual and semiautomatic segmentation were compared using 3 different metric criteria: Dice coefficient, Hausdorff distance, and mean absolute distance. A comparison also was performed between volumes obtained from semiautomatic and manual methods. RESULTS: Manual delineation processing time of tumors for each patient was dependent on its size and complexity and had a mean (±SD) of 12.33 ± 2.47 minutes, whereas it was 3.254 ± 1.7507 minutes for the semiautomatic method. Means of Dice coefficient, Hausdorff distance, and mean absolute distance between manual contours were 0.84 ± 0.02, 2.05 ± 0.66 cm, and 0.78 ± 0.15 cm, and they were 0.82 ± 0.03, 1.91 ± 0.65 cm, and 0.7 ± 0.22 cm between manual and semiautomatic contours, respectively. Moreover, the mean volume ratio (=semiautomatic/manual) calculated for all samples was 0.87. CONCLUSIONS: Given the deformability of this method, the results showed reasonable accuracy and similarity to the results of manual contouring by the oncologists. This study shows that the localized region-based algorithms can have great ability in determining the CTV and can be appropriate alternatives for manual approaches in brain cancer.

Automatic left ventricle segmentation in volumetric SPECT data set by variational level set.

INTRODUCTION: Left ventricle (LV) quantification in nuclear medicine images is a challenging task for myocardial perfusion scintigraphy. A hybrid method for left ventricle myocardial border extraction in SPECT datasets was developed and tested to automate LV ventriculography. METHODS: Automatic segmentation of the LV in volumetric SPECT data was implemented using a variational level set algorithm. The method consists of two steps: (1) initialization and (2) segmentation. Initially, we estimate the initial closed curves in SPECT images using adaptive thresholding and morphological operations. Next, we employ the initial closed curves to estimate the final contour by variational level set. The performance of the proposed approach was evaluated by comparing manually obtained boundaries with automated segmentation contours in 10 SPECT data sets obtained from adult patients. Segmented images by proposed methods were visually compared with manually outlined contours and the performance was evaluated using ROC analysis. RESULTS: The proposed method and a traditional level set method were compared by computing the sensitivity and specificity of ventricular outlines as well as ROC analysis. The results show that the proposed method can effectively segment LV regions with a sensitivity and specificity of 88.9 and 96.8%, respectively. Experimental results demonstrate the effectiveness and reasonable robustness of the automatic method. CONCLUSION: A new variational level set technique was able to automatically trace the LV contour in cardiac SPECT data sets, based on the characteristics of the overall region of LV images. Smooth and accurate LV contours were extracted using this new method, reducing the influence of nearby interfering structures including a hypertrophied right ventricle, hepatic or intestinal activity, and pulmonary or intramammary activity.