Automated glioma grading on conventional MRI images using deep convolutional neural networks.

PURPOSE: Gliomas are the most common primary tumor of the brain and are classified into grades I-IV of the World Health Organization (WHO), based on their invasively histological appearance. Gliomas grading plays an important role to determine the treatment plan and prognosis prediction. In this study we propose two novel methods for automatic, non-invasively distinguishing low-grade (Grades II and III) glioma (LGG) and high-grade (grade IV) glioma (HGG) on conventional MRI images by using deep convolutional neural networks (CNNs). METHODS: All MRI images have been preprocessed first by rigid image registration and intensity inhomogeneity correction. Both proposed methods consist of two steps: (a) three-dimensional (3D) brain tumor segmentation based on a modification of the popular U-Net model; (b) tumor classification on segmented brain tumor. In the first method, the slice with largest area of tumor is determined and the state-of-the-art mask R-CNN model is employed for tumor grading. To improve the performance of the grading model, a two-dimensional (2D) data augmentation has been implemented to increase both the amount and the diversity of the training images. In the second method, denoted as 3DConvNet, a 3D volumetric CNNs is applied directly on bounding image regions of segmented tumor for classification, which can fully leverage the 3D spatial contextual information of volumetric image data. RESULTS: The proposed schemes were evaluated on The Cancer Imaging Archive (TCIA) low grade glioma (LGG) data, and the Multimodal Brain Tumor Image Segmentation (BraTS) Benchmark 2018 training datasets with fivefold cross validation. All data are divided into training, validation, and test sets. Based on biopsy-proven ground truth, the performance metrics of sensitivity, specificity, and accuracy are measured on the test sets. The results are 0.935 (sensitivity), 0.972 (specificity), and 0.963 (accuracy) for the 2D Mask R-CNN based method, and 0.947 (sensitivity), 0.968 (specificity), and 0.971 (accuracy) for the 3DConvNet method, respectively. In regard to efficiency, for 3D brain tumor segmentation, the program takes around ten and a half hours for training with 300 epochs on BraTS 2018 dataset and takes only around 50 s for testing of a typical image with a size of 160 × 216 × 176. For 2D Mask R-CNN based tumor grading, the program takes around 4 h for training with around 60 000 iterations, and around 1 s for testing of a 2D slice image with size of 128 × 128. For 3DConvNet based tumor grading, the program takes around 2 h for training with 10 000 iterations, and 0.25 s for testing of a 3D cropped image with size of 64 × 64 × 64, using a DELL PRECISION Tower T7910, with two NVIDIA Titan Xp GPUs. CONCLUSIONS: Two effective glioma grading methods on conventional MRI images using deep convolutional neural networks have been developed. Our methods are fully automated without manual specification of region-of-interests and selection of slices for model training, which are common in traditional machine learning based brain tumor grading methods. This methodology may play a crucial role in selecting effective treatment options and survival predictions without the need for surgical biopsy.

Three-dimensional alginate hydrogels for radiobiological and metabolic studies of cancer cells.

The purpose of this study is to demonstrate calcium alginate hydrogels as a system for in vitro radiobiological and metabolic studies of cancer cells. Previous studies have established calcium alginate as a versatile three-dimensional (3D) culturing system capable of generating areas of oxygen heterogeneity and modeling metabolic changes in vitro. Here, through dosimetry, clonogenic and viability assays, and pimonidazole staining, we demonstrate that alginate can model radiobiological responses that monolayer cultures do not simulate. Notably, alginate hydrogels with radii greater than 500 µm demonstrate hypoxic cores, while smaller hydrogels do not. The size of this hypoxic region correlates with hydrogel size and improved cell survival following radiation therapy. Hydrogels can also be utilized in hyperpolarized magnetic resonance spectroscopy and extracellular flux analysis. Alginate therefore offers a reproducible, consistent, and low-cost means for 3D culture of cancer cells for radiobiological studies that simulates important in vivo parameters such as regional hypoxia and enables long-term culturing and in vitro metabolic studies.

Brain tumor segmentation using holistically nested neural networks in MRI images.

PURPOSE: Gliomas are rapidly progressive, neurologically devastating, largely fatal brain tumors. Magnetic resonance imaging (MRI) is a widely used technique employed in the diagnosis and management of gliomas in clinical practice. MRI is also the standard imaging modality used to delineate the brain tumor target as part of treatment planning for the administration of radiation therapy. Despite more than 20 yr of research and development, computational brain tumor segmentation in MRI images remains a challenging task. We are presenting a novel method of automatic image segmentation based on holistically nested neural networks that could be employed for brain tumor segmentation of MRI images. METHODS: Two preprocessing techniques were applied to MRI images. The N4ITK method was employed for correction of bias field distortion. A novel landmark-based intensity normalization method was developed so that tissue types have a similar intensity scale in images of different subjects for the same MRI protocol. The holistically nested neural networks (HNN), which extend from the convolutional neural networks (CNN) with a deep supervision through an additional weighted-fusion output layer, was trained to learn the multiscale and multilevel hierarchical appearance representation of the brain tumor in MRI images and was subsequently applied to produce a prediction map of the brain tumor on test images. Finally, the brain tumor was obtained through an optimum thresholding on the prediction map. RESULTS: The proposed method was evaluated on both the Multimodal Brain Tumor Image Segmentation (BRATS) Benchmark 2013 training datasets, and clinical data from our institute. A dice similarity coefficient (DSC) and sensitivity of 0.78 and 0.81 were achieved on 20 BRATS 2013 training datasets with high-grade gliomas (HGG), based on a two-fold cross-validation. The HNN model built on the BRATS 2013 training data was applied to ten clinical datasets with HGG from a locally developed database. DSC and sensitivity of 0.83 and 0.85 were achieved. A quantitative comparison indicated that the proposed method outperforms the popular fully convolutional network (FCN) method. In terms of efficiency, the proposed method took around 10 h for training with 50,000 iterations, and approximately 30 s for testing of a typical MRI image in the BRATS 2013 dataset with a size of 160 × 216 × 176, using a DELL PRECISION workstation T7400, with an NVIDIA Tesla K20c GPU. CONCLUSIONS: An effective brain tumor segmentation method for MRI images based on a HNN has been developed. The high level of accuracy and efficiency make this method practical in brain tumor segmentation. It may play a crucial role in both brain tumor diagnostic analysis and in the treatment planning of radiation therapy.

Output factor comparison of Monte Carlo and measurement for Varian TrueBeam 6 MV and 10 MV flattening filter-free stereotactic radiosurgery system.

The dose measurements of the small field sizes, such as conical collimators used in stereotactic radiosurgery (SRS), are a significant challenge due to many factors including source occlusion, detector size limitation, and lack of lateral electronic equilibrium. One useful tool in dealing with the small field effect is Monte Carlo (MC) simulation. In this study, we report a comparison of Monte Carlo simulations and measurements of output factors for the Varian SRS system with conical collimators for energies of 6 MV flattening filter-free (6 MV) and 10 MV flattening filter-free (10 MV) on the TrueBeam accelerator. Monte Carlo simulations of Varian's SRS system for 6 MV and 10 MV photon energies with cones sizes of 17.5 mm, 15.0 mm, 12.5 mm, 10.0 mm, 7.5 mm, 5.0 mm, and 4.0 mm were performed using EGSnrc (release V4 2.4.0) codes. Varian's version-2 phase-space files for 6 MV and 10 MV of TrueBeam accelerator were utilized in the Monte Carlo simulations. Two small diode detectors Edge (Sun Nuclear) and Small Field Detector (SFD) (IBA Dosimetry) were applied to measure the output factors. Significant errors may result if detector correction factors are not applied to small field dosimetric measurements. Although it lacked the machine-specific kfclin,fmsrQclin,Qmsr correction factors for diode detectors in this study, correction factors were applied utilizing published studies conducted under similar conditions. For cone diameters greater than or equal to 12.5 mm, the differences between output factors for the Edge detector, SFD detector, and MC simulations are within 3.0% for both energies. For cone diameters below 12.5 mm, output factors differences exhibit greater variations.

Patient-specific CT dosimetry calculation: a feasibility study.

Current estimation of radiation dose from computed tomography (CT) scans on patients has relied on the measurement of Computed Tomography Dose Index (CTDI) in standard cylindrical phantoms, and calculations based on mathematical representations of "standard man". Radiation dose to both adult and pediatric patients from a CT scan has been a concern, as noted in recent reports. The purpose of this study was to investigate the feasibility of adapting a radiation treatment planning system (RTPS) to provide patient-specific CT dosimetry. A radiation treatment planning system was modified to calculate patient-specific CT dose distributions, which can be represented by dose at specific points within an organ of interest, as well as organ dose-volumes (after image segmentation) for a GE Light Speed Ultra Plus CT scanner. The RTPS calculation algorithm is based on a semi-empirical, measured correction-based algorithm, which has been well established in the radiotherapy community. Digital representations of the physical phantoms (virtual phantom) were acquired with the GE CT scanner in axial mode. Thermoluminescent dosimeter (TLDs) measurements in pediatric anthropomorphic phantoms were utilized to validate the dose at specific points within organs of interest relative to RTPS calculations and Monte Carlo simulations of the same virtual phantoms (digital representation). Congruence of the calculated and measured point doses for the same physical anthropomorphic phantom geometry was used to verify the feasibility of the method. The RTPS algorithm can be extended to calculate the organ dose by calculating a dose distribution point-by-point for a designated volume. Electron Gamma Shower (EGSnrc) codes for radiation transport calculations developed by National Research Council of Canada (NRCC) were utilized to perform the Monte Carlo (MC) simulation. In general, the RTPS and MC dose calculations are within 10% of the TLD measurements for the infant and child chest scans. With respect to the dose comparisons for the head, the RTPS dose calculations are slightly higher (10%-20%) than the TLD measurements, while the MC results were within 10% of the TLD measurements. The advantage of the algebraic dose calculation engine of the RTPS is a substantially reduced computation time (minutes vs. days) relative to Monte Carlo calculations, as well as providing patient-specific dose estimation. It also provides the basis for a more elaborate reporting of dosimetric results, such as patient specific organ dose volumes after image segmentation.

Dynamic adsorption properties of n-alkyl glucopyranosides determine their ability to inhibit cytolysis mediated by acoustic cavitation.

Suspensions of human leukemia (HL-60) cells readily undergo cytolysis when exposed to ultrasound above the acoustic cavitation threshold. However, n-alkyl glucopyranosides (hexyl, heptyl, and octyl) completely inhibit ultrasound-induced (1057 kHz) cytolysis (Sostaric, et al. Free Radical Biol. Med. 2005, 39, 1539-1548). The efficacy of protection from ultrasound-induced cytolysis was determined by the n-alkyl chain length of the glucopyranosides, indicating that protection efficacy depended on adsorption of n-alkyl glucopyranosides to the gas/solution interface of cavitation bubbles and/or the lipid membrane of cells. The current study tests the hypothesis that "sonoprotection" (i.e., protection of cells from ultrasound-induced cytolysis) in vitro depends on the adsorption of glucopyranosides at the gas/solution interface of cavitation bubbles. To test this hypothesis, the effect of ultrasound frequency (from 42 kHz to 1 MHz) on the ability of a homologous series of n-alkyl glucopyranosides to protect cells from ultrasound-induced cytolysis was investigated. It is expected that ultrasound frequency will affect sonoprotection ability since the nature of the cavitation bubble field will change. This will affect the relative importance of the possible mechanisms for ultrasound-induced cytolysis. Additionally, ultrasound frequency will affect the lifetime and rate of change of the surface area of cavitation bubbles, hence the dynamically controlled adsorption of glucopyranosides to their surface. The data support the hypothesis that sonoprotection efficiency depends on the ability of glucopyranosides to adsorb at the gas/solution interface of cavitation bubbles.

Mechanism of the protective effects of long chain n-alkyl glucopyranosides against ultrasound-induced cytolysis of HL-60 cells.

Recently it has been shown that long chain (C5-C8) n-alkyl glucopyranosides completely inhibit ultrasound-induced cytolysis [J.Z. Sostaric, N. Miyoshi, P. Riesz, W.G. DeGraff, and J.B. Mitchell, Free Radical Biol. Med., 39 (2005) 1539]. This protective effect has possible applications in HIFU (high intensity focused ultrasound) for tumor treatment, and in ultrasound assisted drug delivery and gene therapy. n-Alkyl glucopyranosides with hexyl (5mM), heptyl (3mM), octyl (2mM) n-alkyl chains protected 100% of HL-60 cells in vitro from 1.057 MHz ultrasound-induced cytolysis under a range of conditions that resulted in 35-100% cytolysis in the absence of glucopyranosides. However the hydrophilic methyl-beta-d-glucopyranoside did not protect cells. The surface active n-alkyl glucopyranosides accumulate at the gas-liquid interface of cavitation bubbles. The OH radicals and H atoms formed in collapsing cavitation bubbles react by H-atom abstraction from either the n-alkyl chain or the glucose moiety of the n-alkyl glucopyranosides. Owing to the high concentration of the long chain surfactants at the gas-liquid interface of cavitation bubbles, the initially formed carbon radicals on the alkyl chains are transferred to the glucose moieties to yield radicals which react with oxygen leading to the formation of hydrogen peroxide. In this work, we find that the sonochemically produced hydrogen peroxide yields from oxygen-saturated solutions of long chain (hexyl, octyl) n-alkyl glucopyranosides at 614 kHz and 1.057 MHz ultrasound increase with increasing n-alkyl glucopyranoside concentration but are independent of concentration for methyl-beta-D-glucopyranoside. These results are consistent with the previously proposed mechanism of sonoprotection [J.Z. Sostaric, N. Miyoshi, P. Riesz, W.G. DeGraff, and J.B. Mitchell, Free Radical Biol. Med., 39 (2005) 1539]. This sequence of events prevents sonodynamic cell killing by initiation of lipid peroxidation chain reactions in cellular membranes by peroxyl and/or alkoxyl radicals [V. Misik, P. Riesz, Ann. N.Y. Acad. Sci., 899 (2000) 335].

In vitro microscopic imaging of enhanced thrombolysis with 120-kHz ultrasound in a human clot model.

Substantial enhancement of recombinant tissue plasminogen activator (rt-PA) thrombolysis can be achieved with ultrasound, suggesting its use as an adjunctive treatment in thrombolytic therapy for stroke. A microscopic visualization method was used to measure the lysis of human whole-blood clots treated with human fresh frozen plasma (HFFP), rt-PA, and 120-kHz ultrasound for 30 min at T = 37 ° C. The clot-plasma interface was imaged using an inverted optical microscope and the thrombolytic front analyzed as a function of time. Ultrasound treatment significantly enhanced the mean lytic rate from 0.5 to 3.4 µm/min (a 580% change) compared with rt-PA treatment alone.