In radiomics studies, researchers usually need to develop a supervised machine learning model map image features onto the clinical conclusion. A classical pipeline consists of several steps, including normalization, feature selection, and classification. It is often tedious find an optimal with appropriate combinations. We designed open-source software package named FeAture Explorer (FAE). was programmed Python used NumPy, pandas, scikit-learning modules. FAE can be extract features,...

In this letter a low-drift monocular SLAM method is proposed targeting indoor scenarios, where often fails due to the lack of textured surfaces. Our approach decouples rotation and translation estimation tracking process reduce long-term drift in environments. order take full advantage available geometric information scene, surface normals are predicted by convolutional neural network from each input RGB image real-time. First, drift-free estimated based on lines using spherical mean-shift...

We propose a novel approach aimed at object and semantic scene completion from partial scan represented as 3D point cloud. Our architecture relies on three layers that are used successively within an encoder-decoder structure specifically developed for the task hand. The first one carries out feature extraction by matching features to set of pre-trained local descriptors. Then, avoid losing individual descriptors part standard operations such max-pooling, we alternative neighbor-pooling...

We propose a novel model for 3D semantic completion from single depth image, based on encoder and three separate generators used to reconstruct different geometric representations of the original completed scene, all sharing same latent space. To transfer information between branches network, we introduce paths them concatenating features at corresponding network layers. Motivated by limited amount training samples real scenes, an interesting attribute our architecture is capacity supplement...

Prediction of isocitrate dehydrogenase (IDH) mutation status and epilepsy occurrence are important to glioma patients. Although machine learning models have been constructed for both issues, the correlation between them has not explored. Our study aimed exploit this improve performance IDH identification diagnosis in patients with II-IV. 399 were retrospectively enrolled divided into a training (n = 279) an independent test 120) cohort. Multi-center dataset 228) from The Cancer Imaging...

We propose a method to reconstruct, complete and semantically label 3D scene from single input depth image. improve the accuracy of regressed semantic maps by novel architecture based on adversarial learning. In particular, we suggest using multiple loss terms that not only enforce realistic outputs with respect ground truth, but also an effective embedding internal features. This is done correlating latent features encoder working partial 2.5D data extracted variational auto-encoder trained...

Abstract We propose a novel convolutional operator for the task of point cloud completion. One striking characteristic our approach is that, conversely to related work it does not require any max-pooling or voxelization operation. Instead, proposed used learn embedding in encoder extracts permutation-invariant features from via soft-pooling feature activations, which are able preserve fine-grained geometric details. These then passed on decoder architecture. Due compression encoder, typical...

This study was conducted in order to investigate the association between radiomics features and frontal glioma-associated epilepsy (GAE) propose a reliable radiomics-based model predict GAE. retrospective consecutively enrolled 166 adult patients with glioma (111 training cohort 55 testing cohort). A total 1,130 were extracted from T2 fluid-attenuated inversion recovery images, including first-order statistics, 3D shape, texture, wavelet features. Regions of interest, entire tumor...

To investigate the association between clinic-radiological features and glioma-associated epilepsy (GAE), we developed validated a radiomics nomogram for predicting GAE in WHO grade II~IV gliomas.This retrospective study consecutively enrolled 380 adult patients with glioma (266 training cohort 114 testing cohort). Regions of interest, including entire tumor peritumoral edema, were drawn manually. The semantic radiological characteristics assessed by radiologist 15 years experience...

Most modern deep learning-based multi-view 3D reconstruction techniques use RNNs or fusion modules to combine information from multiple images after independently encoding them. These two separate steps have loose connections and do not allow easy sharing among views. We propose LegoFormer, a transformer model for voxel-based that uses the attention layers share views during all computational stages. Moreover, instead of predicting each voxel independently, we parametrize output with series...

To introduce a new implementation of radiomics analysis for cartilage and subchondral bone the knee to compare performance proposed models classic T2 relaxation time in distinguishing knees predisposed posttraumatic osteoarthritis (PTOA) after anterior cruciate ligament reconstruction (ACLR) healthy controls.114 patients following ACLR at least 2 years 43 controls were reviewed allocated training (n = 110) testing 47) cohorts. Radiomics are built regions different compartments: lateral femur...

Abstract Background Epithelial ovarian cancer (EOC) is the most malignant gynecological tumor in women. This study aimed to construct and compare radiomics-clinical nomograms based on MR images EOC prognosis prediction. Methods A total of 186 patients with pathologically proven were enrolled randomly divided into a training cohort ( n = 130) validation 56). Clinical characteristics each patient retrieved from hospital information system. 1116 radiomics features extracted body T2-weighted...

The diagnosis of prenatal placenta accreta spectrum (PAS) with magnetic resonance imaging (MRI) is highly dependent on radiologists' experience. A deep learning (DL) method using the prior knowledge that PAS-related signs are generally found along utero-placental borderline (UPB) may help radiologists, especially those less experience, to mitigate this issue.

We propose an approach to estimate the 3D pose of a human hand while grasping objects from single RGB image. Our is based on probabilistic model implemented with deep architectures, which used for regressing, respectively, 2D joints heat maps and coordinates. train our networks so make robust large objectand self-occlusions, as commonly occurring task at hand. Using specialized latent variables, architecture internally infers category grasped object enhance reconstruction, underlying...

There have been numerous recently proposed methods for monocular depth prediction (MDP) coupled with the equally rapid evolution of benchmarking tools. However, we argue that MDP is currently witnessing benchmark over-fitting and relying on metrics are only partially helpful to gauge usefulness predictions 3D applications. This limits design development novel truly aware - improving towards estimating structure scene rather than optimizing 2D-based distances. In this work, aim bring...

Lidar became an important component of the perception systems in autonomous driving. But challenges training data acquisition and annotation made emphasized role sensor to domain adaptation. In this letter, we address problem lidar upsampling. Learning on point clouds is rather a challenging task due their irregular sparse structure. Here propose method for cloud upsampling which can reconstruct fine-grained scan patterns. The key idea utilize edge-aware dense convolutions both feature...

CNN has shown excellent performance on object recognition based huge amount of real images. For training with synthetic data rendered from 3D models alone to reduce the workload collecting images, we propose a concatenated self-restraint learning structure lead by triplet and softmax jointed loss function for recognition. Locally connected auto encoder trained images without background used reconstruction against environment variables produces an additional channel automatically RGB channels...