Training Patches Research Articles

Adaptive content-aware spatial regularized correlation filter tracking algorithm

In order to solve the annoying boundary effects in correlation filter (CF) trackers induced by cyclic shift when sampling training patches, and improve the tracking performance, an adaptive content aware spatially regularized correlation filter (ACSRCF) is proposed. Firstly, real negative samples are generated from the background area around the target object, so as to alleviate the filter degradation by the fake negative samples induced from the circularly shifted object patches. Secondly, the locality sensitive histogram (LSH) based foreground feature is extracted and incorporated with the spatial regularization weight which is updated adaptively according to the varied object-oriented appearances. Thereafter, the CF model is optimized using the alternative direction method of multipliers (ADMM) in which the model is decomposed into two sub-problems and the LSH-based features are used in iteration for obtaining the optimal solutions. Finally, the proposed method is evaluated on 5 public benchmarks. The experimental results show that the accuracy and success rate scores of our method on OTB 50 dataset are 90.3%and 66.1%, respectively, exceeding the other CF trackers .The data on the OTB100 dataset is 92.2%and 69.2%, and the accuracy first ranks among all the trackers, while the success rate is ahead of other CF trackers.

The Assessment of Different Calibration Strategies on the Luminance Measurement Using High Dynamic Range Images

ABSTRACT High Dynamic Range Imaging (HDRI) is an increasingly popular technique for studying the lighting environment through luminance maps. However, a thorough calibration process is necessary to retrieve accurate photometric measurements when generating HDR images using commercially affordable equipment like cameras. This paper aims to provide an overview of the most commonly used luminance calibration strategies and evaluate their effectiveness in luminance measurements. Seven strategies were examined in two experiments consisting of four scenes with varying levels of lighting disparity on the training patches and the testing patches. The LDR images were taken by two professional digital single-lens reflex cameras fitted with the standard zoom lens and merged into HDR images using the software Photomatix. The results indicate the much simpler sRG B D 65 method using a reliable standard target as the reference could deliver a reasonable luminance measurement accuracy which is within 10% even for scenes with large luminance difference on the training and testing patches. Although the more complex camera-specific CTM methods can enhance colorimetric accuracy, they do not significantly improve photometric accuracy. Moreover, it was demonstrated that a linear regression model with an intercept, utilizing pixel luminance extracted from multiple grayscale patches in the HDR image, can serve as a potential solution for checking the significance of the scattering effect and providing a reference for the selection of appropriate reference standards. For lighting researchers seeking an appropriate strategy to calibrate HDR luminance maps, this study can serve as a valuable reference.

Deep learning model with pathological knowledge for detection of colorectal neuroendocrine tumor

Colorectal neuroendocrine tumors (NETs) differ significantly from colorectal carcinoma (CRC) in terms of treatment strategy and prognosis, necessitating a cost-effective approach for accurate discrimination. Here, we propose an approach for distinguishing between colorectal NET and CRC based on pathological images by utilizing pathological prior information to facilitate the generation of robust slide-level features. By calculating the similarity between morphological descriptions and patches, our approach selects only 2% of the diagnostically relevant patches for both training and inference, achieving an area under the receiver operating characteristic curve (AUROC) of 0.9974 on the internal dataset, and AUROCs of 0.9724 and 0.9513 on two external datasets. Our model effectively identifies NETs from CRCs, reducing unnecessary immunohistochemical tests and enhancing the precise treatment for patients with colorectal tumors. Our approach also enables researchers to investigate methods with high accuracy and low computational complexity, thereby advancing the application of artificial intelligence in clinical settings.

An algorithm for two-dimensional pattern detection by combining Echo State Network-based weak classifiers

Pattern detection is one of the essential technologies in computer vision. To solve pattern detection problems, the system needs a vast amount of computational resources. To train a multilayer perceptron or convolutional neural network, the gradient descent method is commonly used. The method consumes computational resources. To reduce the amount of computation, we propose a two-dimensional pattern detection algorithm based on Echo State Network (ESN). The training rule of ESN is based on one-shot ridge regression, which enables us to avoid the gradient descent. ESN is a kind of recurrent neural network (RNN), which is often used to embed temporal signals inside the network, rarely used for the embedding of static patterns. In our prior work (Kage, 2023), we found that static patterns can be embedded in an ESN network by associating the training patterns with its stable states, or attractors. By using the same training procedure as our prior work, we made sure that we can associate each training patch image with the desired output vector. The resulting performance of a single ESN classifier is, however, relatively poor. To overcome this poor performance, we introduced an ensemble learning framework by combining multiple ESN weak classifiers. To evaluate the performance, we used CMU-MIT frontal face images (CMU DB). We trained eleven ESN-based classifiers by using six CMU DB training images and evaluated the performance by using a CMU DB test image. We succeeded in reducing false positives in the CMU DB test image down to 0.0515 %.

Hyperparameter tuning for deep learning semantic image segmentation of micro computed tomography scanned fiber-reinforced composites

Image segmentation with deep learning models has significantly improved the accuracy of the pixel-wise labeling of scientific imaging which is critical for many quantitative image analyses. This has been feasible through U-Net and related architecture convolutional neural network models. Although the adoption of these models has been widespread, their training data pool and hyperparameters have been mostly determined by educated guesses through trial and error. In this study, we present observations of how training data volume, data augmentation, and patch size affect deep learning performance within a limited data set. Here we study U-Net model training on four different samples of x-ray CT images of fiber-reinforced composites. Because the training process is not deterministic, we relied on seven-fold replication of each experimental condition to avoid under-sampling and observe model training variance. Unsurprisingly, we find greater training data volume strongly benefits individual models’ final accuracy and learning speed while depressing variance among replicates. Importantly, data augmentation has a profound benefit to model performance, especially in cases with a low abundance of ground truth, and we conclude that high coefficients of data augmentation should be used in scientific imaging semantic segmentation models. Future work to describe and measure image complexity is warranted and likely to ultimately guide researchers on the minimum required training data volume for particular scientific imaging deep learning tasks.

A hybrid data fusion approach with twin CNN architecture for enhancing image source identification in IoT environment

AbstractWith the proliferation of digital devices in internet of things (IoT) environment featuring advanced visual capabilities, the task of Image Source Identification (ISI) has become increasingly vital for legal purposes, ensuring the verification of image authenticity and integrity, as well as identifying the device responsible for capturing the original scene. Over the past few decades, researchers have employed both traditional and machine‐learning methods to classify image sources. In the current landscape, data‐driven approaches leveraging deep learning models have emerged as powerful tools for achieving higher accuracy and precision in source prediction. The primary focus of this research is to address the complexities arising from diverse image sources and variable quality in IoT‐generated multimedia data. To achieve this, a Hybrid Data Fusion Approach is introduced, leveraging multiple sources of information to bolster the accuracy and robustness of ISI. This fusion methodology integrates diverse data streams from IoT devices, including metadata, sensor information, and contextual data, amalgamating them into a comprehensive data set for analysis. This study introduces an innovative approach to ISI through the implementation of a Twin Convolutional Neural Network Architecture (TCA) aimed at enhancing the efficacy of source classification. In TCA, the first CNN architecture, referred to as DnCNN, is employed to eliminate noise from the original data set, generating 256 × 256 patches for both training and testing. Subsequently, the second CNN architecture is employed to classify images based on features extracted from various convolutional layers using a 3 × 3 filter, thereby enhancing prediction efficiency. The proposed model demonstrates exceptional accuracy in effectively classifying image sources, showcasing its potential as a robust solution in the realm of ISI.

A deep learning framework for suppressing prestack seismic random noise without noise-free labels

Random noise attenuation is significant in seismic data processing. Supervised deep learning-based denoising methods have been widely developed and applied in recent years. In practice, it is often time-consuming and laborious to obtain noise-free data for supervised learning. Therefore, we propose a novel deep learning framework to denoise prestack seismic data without clean labels, which trains a high-resolution residual neural network (SRResnet) with noisy data for input and the same valid data with different noise for output. Since valid signals in noisy sample pairs are spatially correlated and random noise is spatially independent and unpredictable, the model can learn the features of valid data while suppressing random noise. Noisy data targets are generated by a simple conventional method without fine-tuning parameters. The initial estimates allow signal or noise leakage as the network does not require clean labels. The Monte Carlo strategy is applied to select training patches for increasing valid patches and expanding training datasets. Transfer learning is used to improve the generalization of real data processing. The synthetic and real data tests perform better than the commonly used state-of-the-art denoising methods.

Large-Scale Spatial–Temporal Identification of Urban Vacant Land and Informal Green Spaces Using Semantic Segmentation

The large-scale identification of urban vacant land (UVL) and informal green spaces (IGSs) using conventional identification methods is challenged by the high cost of time and resources as well as inconsistent outcomes. Moreover, the spatial–temporal changes of UVL and IGSs have received limited academic attention. We introduce a methodological framework for the large-scale automatic identification of UVL and IGSs in Hangzhou, China, based on semantic segmentation. We construct and release a large-scale dataset for UVL identification, containing five different UVL categories, and one of them is IGS, with 3096 patches for training and 128 patches for evaluation. We then train five different semantic segmentation networks using the dataset and utilise Segformer to predict UVL and IGSs within the whole urban area of Hangzhou. The presented segmentation model has a hierarchically structured Transformer encoder and a multilayer perceptron decoder, which incorporates local and global information to obtain effective feature representations. Extensive experiments have been conducted to evaluate the segmentation performance of the applied Segformer. Experimental results show the good identification performance of Segformer for UVL and IGSs. Results also verify that the proposed identification framework can be effectively used to analyse the spatial–temporal changes of UVL and IGSs in Hangzhou.

Deep learning with soft attention mechanism for small-scale ground roll attenuation

Ground roll is a type of coherent noise with low frequency, low velocity, and high amplitude, which masks useful signals and decreases the quality of subsequent seismic data processing. It is a challenge for traditional signal processing methods to separate useful signals effectively when the ground roll and useful reflected signals overlap seriously in the low-frequency band. We develop a supervised-learning-based framework with soft attention residual learning mechanisms for suppressing the ground roll noise. To reduce the cost of manual labeling, the 2D patching technique is used to segment large-scale seismic data into a large number of small-scale patches for training. Our network includes a multibranch attention block that uses multiple branches with different kernel sizes to extract waveform features at different scales from input noisy patches. Then, we use the soft attention mechanism to select and fuse the feature maps of different branches. Our network can achieve encouraging ground roll attenuation performance by using a small number of training samples, which is demonstrated by synthetic and field data examples. Compared with one traditional method and two advanced deep-learning frameworks, our network has better abilities in preserving low-frequency useful signals and removing ground roll.

Facies conditional simulation based on VAE-GAN model and image quilting algorithm

Characterization of complex reservoir structures by using limited observations is challenging in geosciences because it requires to reproduce geological realism. We propose a novel method to reconstruct complex structures by combining variational autoencoder and generative adversarial networks (VAE-GAN) with conditional image quilting algorithm. It improves the stability of traditional GAN-based simulation method without decreasing the quality of patterns. Firstly, we construct a VAE-GAN model to extract the high-dimensional features of facies patterns and to create abundant new patterns. The VAE-GAN-based learning method has a good ability in feature learning from training patches and accurately reproduce new facies patterns. Finally, these new patterns are spliced together to reconstruct a complex geological structure by employing the conditional image quilting algorithm, in which patterns are pasted to the simulated areas based on the calculated minimum cost path. During this process, conditioning is also considered. Since the pattern is generated by using the deep learning method rather than directly extracted from the training image, the diversity of realizations is enhanced without losing reproducibility. Based on synthetic training image, several examples are described and analyzed in detail, demonstrating the effectiveness and reliability of the present method. In addition, the new method is applied to a real training image. The complex heterogeneous structures are well reproduced by our method, indicating its practicability.

Improved artificial intelligence discrimination of minor histological populations by supplementing with color-adjusted images

Despite the dedicated research of artificial intelligence (AI) for pathological images, the construction of AI applicable to histopathological tissue subtypes, is limited by insufficient dataset collection owing to disease infrequency. Here, we present a solution involving the addition of supplemental tissue array (TA) images that are adjusted to the tonality of the main data using a cycle-consistent generative adversarial network (CycleGAN) to the training data for rare tissue types. F1 scores of rare tissue types that constitute < 1.2% of the training data were significantly increased by improving recall values after adding color-adjusted TA images constituting < 0.65% of total training patches. The detector also enabled the equivalent discrimination of clinical images from two distinct hospitals and the capability was more increased following color-correction of test data before AI identification (F1 score from 45.2 ± 27.1 to 77.1 ± 10.3, p < 0.01). These methods also classified intraoperative frozen sections, while excessive supplementation paradoxically decreased F1 scores. These results identify strategies for building an AI that preserves the imbalance between training data with large differences in actual disease frequencies, which is important for constructing AI for practical histopathological classification.

SGCL: Spatial guided contrastive learning on whole-slide pathological images.

Self-supervised representation learning (SSL) has achieved remarkable success in its application to natural images while falling behind in performance when applied to whole-slide pathological images (WSIs). This is because the inherent characteristics of WSIs in terms of gigapixel resolution and multiple objects in training patches are fundamentally different from natural images. Directly transferring the state-of-the-art (SOTA) SSL methods designed for natural images to WSIs will inevitably compromise their performance. We present a novel scheme SGCL: Spatial Guided Contrastive Learning, to fully explore the inherent properties of WSIs, leveraging the spatial proximity and multi-object priors for stable self-supervision. Beyond the self-invariance of instance discrimination, we expand and propagate the spatial proximity for the intra-invariance from the same WSI and inter-invariance from different WSIs, as well as propose the spatial-guided multi-cropping for inner-invariance within patches. To adaptively explore such spatial information without supervision, we propose a new loss function and conduct a theoretical analysis to validate it. This novel scheme of SGCL is able to achieve additional improvements over the SOTA pre-training methods on diverse downstream tasks across multiple datasets. Extensive ablation studies have been carried out and visualizations of these results have been presented to aid understanding of the proposed SGCL scheme. As open science, all codes and pre-trained models are available at https://github.com/HHHedo/SGCL.

The use of deep learning state-of-the-art architectures for oral epithelial dysplasia grading: A comparative appraisal.

Dysplasia grading systems for oral epithelial dysplasia are a source of disagreement among pathologists. Therefore, machine learning approaches are being developed to mitigate this issue. This cross-sectional study included a cohort of 82 patients with oral potentially malignant disorders and correspondent 98 hematoxylin and eosin-stained whole slide images with biopsied-proven dysplasia. All whole-slide images were manually annotated based on the binary system for oral epithelial dysplasia. The annotated regions of interest were segmented and fragmented into small patches and non-randomly sampled into training/validation and test subsets. The training/validation data were color augmented, resulting in a total of 81,786 patches for training. The held-out independent test set enrolled a total of 4,486 patches. Seven state-of-the-art convolutional neural networks were trained, validated, and tested with the same dataset. The models presented a high learning rate, yet very low generalization potential. At the model development, VGG16 performed the best, but with massive overfitting. In the test set, VGG16 presented the best accuracy, sensitivity, specificity, and area under the curve (62%, 62%, 66%, and 65%, respectively), associated with the higher loss among all Convolutional Neural Networks (CNNs) tested. EfficientB0 has comparable metrics and the lowest loss among all convolutional neural networks, being a great candidate for further studies. The models were not able to generalize enough to be applied in real-life datasets due to an overlapping of features between the two classes (i.e., high risk and low risk of malignization).

Fast diffraction model of an EUV mask based on asymmetric patch data fitting.

Calculating the diffraction near field (DNF) of a three-dimensional (3D) mask is a key problem in the extreme ultraviolet (EUV) lithography imaging modeling. This paper proposes a fast DNF model of an EUV mask based on the asymmetric patch data fitting method. Due to the asymmetric imaging characteristics of the EUV lithography system, a DNF library is built up including the training mask patches posed in different orientations and their rigorous DNF results. These training patches include some representative local mask features such as the convex corners, concave corners, and edge segments in four directions. Then, a convolution-based compact model is developed to rapidly simulate the DNFs of 3D masks, where the convolution kernels are inversely calculated to fit all of the training data. Finally, the proposed model is verified by simulation experiments. Compared to a state-of-the-art EUV mask model based on machine learning, the proposed method can further reduce the computation time by 60%-70% and roughly obtain the same simulation accuracy.

Pre-Processing Training Data Improves Accuracy and Generalisability of Convolutional Neural Network Based Landscape Semantic Segmentation

Data pre-processing for developing a generalised land use and land cover (LULC) deep learning model using earth observation data is important for the classification of a different date and/or sensor. However, it is unclear how to approach deep learning segmentation problems in earth observation data. In this paper, we trialled different methods of data preparation for Convolutional Neural Network (CNN) training and semantic segmentation of LULC features within aerial photography over the Wet Tropics and Atherton Tablelands, Queensland, Australia. This was conducted by trialling and ranking various training patch selection sampling strategies, patch and batch sizes, data augmentations and scaling and inference strategies. Our results showed: a stratified random sampling approach for producing training patches counteracted class imbalances; a smaller number of larger patches (small batch size) improves model accuracy; data augmentations and scaling are imperative in creating a generalised model able to accurately classify LULC features in imagery from a different date and sensor; and producing the output classification by averaging multiple grids of patches and three rotated versions of each patch produced a more accurate and aesthetic result. Combining the findings from the trials, we fully trained five models on the 2018 training image and applied the model to the 2015 test image. The output LULC classifications achieved an average kappa of 0.84, user accuracy of 0.81, and producer accuracy of 0.87. Future research using CNNs and earth observation data should implement the findings of this project to increase LULC model accuracy and transferability.

Objective evaluation-based efficient learning framework for hyperspectral image classification

ABSTRACT Deep learning techniques with remarkable performance have been successfully applied to hyperspectral image (HSI) classification. Due to the limited availability of training data, earlier studies primarily adopted the patch-based classification framework, which divides images into overlapping patches for training and testing. However, this framework results in redundant computations and possible information leakage. This study proposes an objective evaluation-based efficient learning framework for HSI classification. It consists of two main parts: (i) a leakage-free balanced sampling strategy and (ii) an efficient fully convolutional network (EfficientFCN) optimized for the accuracy-efficiency trade-off. The leakage-free balanced sampling strategy first generates balanced and non-overlapping training and test data by partitioning the HSI and its ground truth image into non-overlapping windows. Then, the generated training and test data are used to train and test the proposed EfficientFCN. EfficientFCN exhibits a pixel-to-pixel architecture with modifications for faster inference speed and improved parameter efficiency. Experimental results demonstrate that the proposed sampling strategy can provide objective performance evaluation. EfficientFCN outperforms many state-of-the-art approaches concerning the speed-accuracy trade-off. For instance, compared to the recent efficient models EfficientNetV2 and ConvNeXt, EfficientFCN achieves 0.92% and 3.42% superior accuracy and 0.19s and 0.16s faster inference time, respectively, on the Houston dataset. Code is available at https://github.com/xmzhang2018.

Corneal endothelial image segmentation training data generation using GANs. Do experts need to annotate?

Objective.Current solutions for corneal endothelial image segmentation use convolutional neural networks. However, their potential is not exploited due to the scarcity of labeled corneal endothelial data caused by an expensive cell delineation process. Therefore, this work proposes synthesizing cell edges and corresponding images using generative adversarial neural networks. To our knowledge, such work has not yet been reported in the context of corneal endothelial image segmentation. Methods.A two-step pipeline for synthesizing training patches is proposed. First, a custom mosaic generator creates a grid that mimics a binary map of endothelial cell edges. The synthetic cell edges are then input to the generative adversarial neural network, which is trained to generate corneal endothelial image patches from the corresponding edge labels. Finally, pairs of synthetic patches are used to train the patch-based U-Net model for corneal endothelial image segmentation. Results.Experiments performed on three datasets of corneal endothelial images show that using purely synthetic data for U-Net training allows image segmentation with comparable accuracy to that obtained when using original images annotated by experts. Depending on the dataset, the segmentation quality decreases only from 1% to 4%. Conclusions:Our solution provides a cost-effective source of diverse training data for corneal endothelial image segmentation. Significance.Due to the simple graphical user interface wrapping the proposed solution, many users can easily obtain unlimited training data for corneal endothelial image segmentation and use it in various scenarios.

Utility of daily 3 m Planet Fusion Surface Reflectance data for tillage practice mapping with deep learning

Tillage practices alter soil surface structure that can be potentially captured by satellite images with both high spatial and temporal resolution. This study explored tillage practice mapping using the daily Planet Fusion surface reflectance (PF-SR) gap-free 3 m data generated by fusing PlanetScope with Landsat-8, Sentinel-2 and MODIS surface reflectance data. The study area is a 220 × 220 km2 agricultural area in South Dakota, USA, and the study used 3285 PF-SR images from September 1, 2020 to August 31, 2021. The PF-SR images for the surveyed 433 fields were sliced into 10,747 training (70%) and evaluation (30%) non-overlapping time series patches. The training and evaluation patches were from different fields for evaluation data independence. The performance of four deep learning models (i.e., 2D convolutional neural networks (CNN), 3D CNN, CNN-Long short-term memory (LSTM), and attention CNN-LSTM) in tillage practice mapping, as well as their sensitivity to different spatial (i.e., 3 m, 24 m, and 96 m) and temporal resolutions (16-day, 8-day, 4-day, 2-day and 1-day) were examined. Classification accuracy continuously increased with increases in both temporal and spatial resolutions. The optimal models (3D CNN and attention CNN-LSTM) achieved ∼77% accuracy using 2-day or daily 3 m resolution data as opposed to ∼72% accuracy using 16-day 3 m resolution data or daily 24 m resolution data. This study also analyzed the feature importance of different acquisition dates for the two optimal models. The 3D CNN model feature importances were found to agree well with the tillage practice time. High feature importance was associated with observations during the fall and spring tillage period (i.e., fresh tillage signals) whereas the crop peak growing period (i.e., tillage signals weathered and confounded by dense canopy) was characterized by a relatively low feature importance. The work provides valuable insights into the utility of deep learning for tillage mapping and change event time identification based on high resolution imagery.

PD38-07 STIMULATED RAMAN HISTOLOGY INTERPRETATION BY ARTIFICIAL INTELLIGENCE CAN PROVIDE REAL-TIME PATHOLOGIC FEEDBACK OF PROSTATE BIOPSIES

You have accessJournal of UrologyCME1 Apr 2023PD38-07 STIMULATED RAMAN HISTOLOGY INTERPRETATION BY ARTIFICIAL INTELLIGENCE CAN PROVIDE REAL-TIME PATHOLOGIC FEEDBACK OF PROSTATE BIOPSIES Miles Mannas, Alec Mitchell, Fang-Ming Deng, Adrian Ion-Margineanu, Derek Jones, Deepthi Hoskoppal, Jonothan Melamed, Steve Pastore, Christian Freudiger, Daniel Orringer, and Samir Taneja Miles MannasMiles Mannas More articles by this author , Alec MitchellAlec Mitchell More articles by this author , Fang-Ming DengFang-Ming Deng More articles by this author , Adrian Ion-MargineanuAdrian Ion-Margineanu More articles by this author , Derek JonesDerek Jones More articles by this author , Deepthi HoskoppalDeepthi Hoskoppal More articles by this author , Jonothan MelamedJonothan Melamed More articles by this author , Steve PastoreSteve Pastore More articles by this author , Christian FreudigerChristian Freudiger More articles by this author , Daniel OrringerDaniel Orringer More articles by this author , and Samir TanejaSamir Taneja More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003336.07AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: There is a delay between prostate biopsy and pathologic diagnosis; this delay has limited the use pathologic feedback during the diagnosis and treatment of prostate cancer (PCa). Stimulated Raman histology (SRH) is a novel microscopic technique allowing real time, label-free, high-resolution microscopic images of unprocessed, un-sectioned tissue. We hypothesized that an artificial intelligence convolutional neural network (CNN) could rapidly interpret prostate biopsy SRH. METHODS: Prostate biopsies were prospectively taken ex-vivo from prostatectomy specimens and scanned in a SRH microscope at 20 microns depth using two Raman shifts: 2845cm-1 and 2930cm-1, to create SRH images. The cores were then processed and H&E-stained as per normal pathologic protocols and used for ground truth pathologic assessment. 303 ex-vivo prostate biopsies taken from 100 radical prostatectomy specimens were used to train the SRH Inception-ResNet-v2 CNN. With a two-sided alpha level of 5%, it was calculated 32 biopsies would provide 90% power to detect a difference in concordance kappa when testing the CNN. Concordance and diagnostic accuracy of the CNN were tested on training and validation patches, and the 32 leave-one out prostate biopsies from 23 radical prostatectomy specimens. RESULTS: The CNN showed a 99.6% weighted accuracy on the training patches and 98.6% when tested on the validation set. The CNN also showed very good kappa concordance (k=0.925, p<0.001) when classifying the 32 prostate biopsies as benign or malignant, giving a diagnostic accuracy of 96.9%, with a scan time of 2-2.75 minutes, Table 1. Additionally, if a region containing PCa was first scanned with SRH, the AI could identify PCa in ∼1 minute. CONCLUSIONS: Artificial intelligence applied to SRH can rapidly classify fresh, unprocessed, unstained prostate biopsies as benign or malignant. Source of Funding: NIH UL1TR001145 © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e995 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Miles Mannas More articles by this author Alec Mitchell More articles by this author Fang-Ming Deng More articles by this author Adrian Ion-Margineanu More articles by this author Derek Jones More articles by this author Deepthi Hoskoppal More articles by this author Jonothan Melamed More articles by this author Steve Pastore More articles by this author Christian Freudiger More articles by this author Daniel Orringer More articles by this author Samir Taneja More articles by this author Expand All Advertisement PDF downloadLoading ...

High-fidelity point cloud completion with low-resolution recovery and noise-aware upsampling

Completing an unordered partial point cloud is a challenging task. Existing approaches that rely on decoding a latent feature to recover the complete shape, often lead to the completed point cloud being over-smoothing, losing details, and noisy. Instead of decoding a whole shape, we propose to decode and refine a low-resolution (low-res) point cloud first, and then perform a patch-wise noise-aware upsampling rather than interpolating the whole sparse point cloud at once, which tends to lose details. Regarding the possibility of lacking details of the initially decoded low-res point cloud, we propose an iterative refinement to recover the geometric details and a symmetrization process to preserve the trustworthy information from the input partial point cloud. After obtaining a sparse and complete point cloud, we propose a patch-wise upsampling strategy. Patch-based upsampling allows to recover fine details better rather than decoding a whole shape. The patch extraction approach is to generate training patch pairs between the sparse and ground-truth point clouds with an outlier removal step to suppress the noisy points from the sparse point cloud. Together with the low-res recovery, our whole pipeline can achieve high-fidelity point cloud completion. Comprehensive evaluations are provided to demonstrate the effectiveness of the proposed method and its components.