Distinct tumor signatures using deep learning-based characterization of the peritumoral microenvironment in glioblastomas and brain metastases

Tumor types are classically distinguished based on biopsies of the tumor itself, as well as a radiological interpretation using diverse MRI modalities. In the current study, the overarching goal is to demonstrate that primary (glioblastomas) and secondary (brain metastases) malignancies can be differentiated based on the microstructure of the peritumoral region. This is achieved by exploiting the extracellular water differences between vasogenic edema and infiltrative tissue and training a convolutional neural network (CNN) on the Diffusion Tensor Imaging (DTI)-derived free water volume fraction. We obtained 85% accuracy in discriminating extracellular water differences between local patches in the peritumoral area of 66 glioblastomas and 40 metastatic patients in a cross-validation setting. On an independent test cohort consisting of 20 glioblastomas and 10 metastases, we got 93% accuracy in discriminating metastases from glioblastomas using majority voting on patches. This level of accuracy surpasses CNNs trained on other conventional DTI-based measures such as fractional anisotropy (FA) and mean diffusivity (MD), that have been used in other studies. Additionally, the CNN captures the peritumoral heterogeneity better than conventional texture features, including Gabor and radiomic features. Our results demonstrate that the extracellular water content of the peritumoral tissue, as captured by the free water volume fraction, is best able to characterize the differences between infiltrative and vasogenic peritumoral regions, paving the way for its use in classifying and benchmarking peritumoral tissue with varying degrees of infiltration.

Distinct Tumor Signatures using Deep Learning-based Characterization of the Peritumoral Microenvironment in Glioblastomas and Brain Metastases

Tumor types are classically distinguished based on biopsies of the tumor itself, as well as a radiological interpretation using diverse MRI modalities. In the current study, the overarching goal is to demonstrate that primary (glioblastomas) and secondary (brain metastases) malignancies can be differentiated based on the microstructure of the peritumoral region. This is achieved by exploiting the extracellular water differences between vasogenic edema and infiltrative tissue and training a convolutional neural network (CNN) on the Diffusion Tensor Imaging (DTI)-derived free water volume fraction map of the peritumoral regions of 54 metastases and 89 glioblastoma patients. We obtained 93% accuracy in discriminating metastases from glioblastomas. This level of accuracy surpasses CNNs trained on other conventional DTI-based measures such as fractional anisotropy (FA) and mean diffusivity (MD), that have been used in other studies. Additionally, the CNN captures the peritumoral heterogeneity better than conventional texture features, including Gabor and radiomic features. Our results demonstrate that the extracellular water content of the peritumoral tissue, as captured by the free water volume fraction, is best able to characterize the differences between infiltrative and vasogenic peritumoral regions, paving the way for its use in classifying and benchmarking peritumoral tissue with varying degrees of infiltration.

Macroscopical monolayer films of ordered arrays of gold nanoparticles as SERS substrates for in-situ quantitative detection in aqueous solutions

In this work, macroscopical monolayer films of ordered arrays of gold nanoparticles (MMF-OA-Au NPs) are successfully prepared at the interfaces of toluene-diethylene glycol (DEG) with a water volume fraction of...

An acylhydrazone-based AIE organogel for the selective sensing of submicromolar level Al3+ and Al(iii)-based metallogel formation to detect oxalic acid

The compound L can be fluorescence-tunable depending on the water volume fraction and optically sense Al3+ without interference.

NIMG-21. DIFFERENTIATING TUMOR TYPES BASED ON THE PERITUMORAL MICROENVIRONMENT USING CONVOLUTIONAL NEURAL NETWORKS

PURPOSE The differential diagnosis of glioblastoma (GBM) versus single brain metastasis (Met) is clinically important, and is undertaken with a clinical reading of MR images and/or tumor biopsy. We investigate whether Mets and GBMs can be differentiated based on the microstructure of the FLAIR-hyperintense peritumoral region measured by diffusion tensor imaging (DTI). We hypothesize that the peritumoral microstructure differs in extracellular water content, based on whether it is vasogenic edema or infiltrative. We use deep learning trained on DTI-based free-water volume fraction maps to discriminate between the peritumoral regions of Met and GBM neoplasms. Our results are also compared with mean diffusivity (MD), the most commonly used DTI metric. METHOD dMRI data of 143 patients with brain tumors (89 glioblastomas and 54 metastases, ages 19-87 years, 77 females and 66 males) were included. Free-water volume fraction maps were computed for the peritumoral regions (demarcated automatically). We developed a 7-layer convolutional neural network (CNN) architecture to distinguish microstructural patterns of Met and GBM tumors using 32 x 32 mm patches placed at random in the peritumoral area. The CNN was trained on patches from a training set of 113 patients and tested on the remaining 30 patients, where majority voting was applied to predict the tumor type for each patient. Although MD has been previously used in both tumor and peritumoral area for discriminating tumor type, we replicated the same process with MD only in the peritumoral area to provide a stronger comparison. RESULT We predicted tumor type with 93% accuracy, outperforming MD with 84% accuracy. CONCLUSION Our results demonstrate that deep learning with CNN on DTI-based free-water volume fraction map can be a promising tool for automatic distinction of tumor types, and has potential as a tumor biomarker.

NIMG-45. DEEP LEARNING-BASED PERITUMORAL MICROSTRUCTURE MAPPING IN GLIOBLASTOMAS USING FREE WATER VOLUME FRACTION

PURPOSE Characterization of the peritumoral microenvironment is a widely researched, but as yet unsolved problem. Determining the tissue microstructure differences between tumor types, arising from differences in infiltration, edema, and disease driven changes in cellularity is important to be able to guide treatment options. Diffusion tensor imaging with characterization of extracellular free water provides unique information of the tissue microstructure. The goal of this work is to leverage this information by applying deep learning on free water maps and create a microstructure map of the peritumoral area, to aid in targeted resection, radiotherapy and treatment management in the form of a crucial supplemental radiomic feature, replacing all other diffusion derived measures. METHOD We leveraged the widely different peritumoral microenvironments of GBMs and Metastatic tumors to create the microstructure maps. Tumor and peritumoral regions were automatically delineated in 143 patients with brain tumors (89 glioblastomas and 54 metastasis, ages 19-87 years, 77 females), and free water maps were computed in the peritumoral regions using their DTI data. We trained a Convolutional Neural Network (CNN) on 32x32 mm patches in the peritumoral area from GBMs and Mets, labeled as non-enhancing tumor (low free-water) and edema (high free-water), respectively. An independent test set was used and the CNN associated a voxel-wise probability to their peritumoral region to produce microstructure maps of GBMs and Mets which were then statistically compared. RESULT For comparison, a t-test was used showing significant group difference in the microstructure map between Mets and GBMs (p< 0.05). CONCLUSION The voxel-wise microstructure map is able to capture the cellularity differences in the peritumoral region, based on DTI-based characterization of the tissue microstructure. CLINICAL IMPORTANCE The microstructure map provides a novel insight into the peritumoral microenvironment using a measure that can be derived from clinically feasible DTI data, replacing pre-existing DTI measures.

Performance Investigation of the Immersed Depth Effects on a Water Wheel Using Experimental and Numerical Analyses

The purpose of this research is to study the effect of different immersed depths on water wheel performance and flow characteristics using numerical simulations. The results indicate that the simulation methods are consistent with experiments with a maximum error less than 5%. Under the same rotational speeds, the efficiency is much higher and the fluctuation amplitude of the torque is much smaller as the immersed radius ratio increases, and until an immersed radius ratio of 82.76%, the wheel shows the best performance, achieving a maximum efficiency of 18.05% at a tip-speed ratio (TSR) of 0.1984. The average difference in water level increases as the immersed radius ratio increases until 82.76%. The water area is much wider and the water volume fraction shows more intense change at the inlet stage at a deep immersed depth. At an immersed radius ratio of 82.76%, some air intrudes into the water at the inlet stage, coupled with a dramatic change in the water volume fraction that would make the flow more complex. Furthermore, eddies are found to gradually generate in a single flow channel nearly at the same time, except for an immersed depth of 1.2 m. However, eddies generate in two flow channels and can develop initial vortexes earlier than other cases because of the elevation of the upstream water level at an immersed radius ratio of 82.76%.