Raw Scan Research Articles

Development of a Machine Learning-Enabled Virtual Reality Tool for Preoperative Planning of Functional Endoscopic Sinus Surgery.

Objectives Virtual reality (VR) is an increasingly valuable teaching tool, but current simulators are not typically clinically scalable due to their reliance on inefficient manual segmentation. The objective of this project was to leverage a high-throughput and accurate machine learning method to automate data preparation for a patient-specific VR simulator used to explore preoperative sinus anatomy. Methods An endoscopic VR simulator was designed in Unity to enable interactive exploration of sinus anatomy. The Saak transform, a data-efficient machine learning method, was adapted to accurately segment sinus computed tomography (CT) scans using minimal training data, and the resulting data were reconstructed into three-dimensional (3D) patient-specific models that could be explored in the simulator. Results Using minimal training data, the Saak transform-based machine learning method offers accurate soft-tissue segmentation. When explored with an endoscope in the VR simulator, the anatomical models generated by the algorithm accurately capture key sinus structures and showcase patient-specific variability in anatomy. Conclusion By offering an automatic means of preparing VR models from a patient's raw CT scans, this pipeline takes a key step toward clinical scalability. In addition to preoperative planning, this system also enables virtual endoscopy-a tool that is particularly useful in the COVID-19 era. As VR technology inevitably continues to develop, such a foundation will help ensure that future innovations remain clinically accessible.

Use of a nationwide, patient-powered observational platform to define the landscape of treatment, molecular profiling, and outcomes for fibrolamellar carcinoma.

e16192 Background: Fibrolamellar carcinoma (FLC) is a rare form of liver cancer, distinct from hepatocellular carcinoma, characterized by its prevalence among adolescents and young adults, capacity to form bulky tumors, and high rate of recurrence and metastasis. Without an established standard-of-care, the treatment of FLC is multidisciplinary, often involving surgery, radiation, and therapeutic interventions concurrently. Understanding the landscape of real-world treatments for FLC pts is imperative to defining an effective therapeutic approach. Methods: In partnership with the FibroFighters Foundation, we enrolled 102 patients (pts) in the observational research protocol XCELSIOR ( NCT03793088 ). Electronic medical records (EMR) were gathered from over 2,000 sites of care across 31 states - a median of 43 unique locations per pt - and utilized to populate a 21 CFR Part 11-compliant database with fully longitudinal, uninterrupted clinical data. Data elements were standardized to OMOP-based dictionaries. Anti-cancer medications were combined to define regimens and correlated with surgical dates. Pts were offered access to aggregated EMR and structured summaries as a digital personal health record via online portals. Results: Median age at diagnosis was 19 years (range 6-51 years) and median diagnosis date was 8/26/2020. At initial diagnosis, 36% of pts had distant metastases, but 86% of all pts eventually experienced metastatic disease with lung being the most frequent location (30 pts). The most prevalent anti-cancer therapies used were lenvatinib (75% pts exposed) and nivolumab (73% pts) and the most prevalent regimen was gemcitabine/lenvatinib/oxaliplatin (33% of pts). Among repurposed agents, 50% were exposed to quercetin, 44% to celecoxib, and 14% to metformin. 64% of pts underwent at least one surgical resection with 21% of these receiving neoadjuvant therapy including 7 distinct regimens, most commonly gemcitabine/lenvatinib/oxaliplatin. 91% of pts received adjuvant therapy for a median duration of 162 days. 22% of pts received radiation therapy and 15% received Y90 radioembolization. Among diagnostic profiling, 82% had somatic next generation sequencing, 48% had germline testing, 28.5% had organoid functional profiling, and 68% had Signatera testing performed. Conclusions: Patients with FLC undergo extensive molecular profiling with multiple novel therapeutic regimens being employed in real world practice. Aggregation of raw imaging scans is ongoing for central radiological review to measure depth and duration of response offered by the recorded therapeutic strategies. This dataset is a rich resource available to academic and government researchers to fuel research and define optimal treatment modalities. Topline statistics for the dataset are available at https://xcures.com/flcdashboard/ . Clinical trial information: NCT03793088 .

Advancing Pulmonary Nodule Diagnosis by Integrating Engineered and Deep Features Extracted from CT Scans

Enhancing lung cancer diagnosis requires precise early detection methods. This study introduces an automated diagnostic system leveraging computed tomography (CT) scans for early lung cancer identification. The main approach is the integration of three distinct feature analyses: the novel 3D-Local Octal Pattern (LOP) descriptor for texture analysis, the 3D-Convolutional Neural Network (CNN) for extracting deep features, and geometric feature analysis to characterize pulmonary nodules. The 3D-LOP method innovatively captures nodule texture by analyzing the orientation and magnitude of voxel relationships, enabling the distinction of discriminative features. Simultaneously, the 3D-CNN extracts deep features from raw CT scans, providing comprehensive insights into nodule characteristics. Geometric features and assessing nodule shape further augment this analysis, offering a holistic view of potential malignancies. By amalgamating these analyses, our system employs a probability-based linear classifier to deliver a final diagnostic output. Validated on 822 Lung Image Database Consortium (LIDC) cases, the system’s performance was exceptional, with measures of 97.84%, 98.11%, 94.73%, and 0.9912 for accuracy, sensitivity, specificity, and Area Under the ROC Curve (AUC), respectively. These results highlight the system’s potential as a significant advancement in clinical diagnostics, offering a reliable, non-invasive tool for lung cancer detection that promises to improve patient outcomes through early diagnosis.

Semantic segmentation of raw multispectral laser scanning data from urban environments with deep neural networks

Real-time semantic segmentation of point clouds has increasing importance in applications related to 3D city modelling and mapping, automated inventory of forests, autonomous driving and mobile robotics. Current state-of-the-art point cloud semantic segmentation methods rely heavily on the availability of 3D laser scanning data. This is problematic in regards of low-latency, real-time applications that use data from high-precision mobile laser scanners, as those are typically 2D line scanning devices. In this study, we experiment with real-time semantic segmentation of high-density multispectral point clouds collected from 2D line scanners in urban environments using encoder - decoder convolutional neural network architectures. We introduce a rasterized multi-scan input format that can be constructed exclusively from the raw (non-georeferenced profiles) 2D laser scanner measurement stream without odometry information. In addition, we investigate the impact of multispectral data on the segmentation accuracy. The dataset used for training, validation and testing was collected with multispectral FGI AkhkaR4-DW backpack laser scanning system operating at the wavelengths of 905 nm and 1550 nm, and consists in total of 228 million points (39 583 scans). The data was divided into 13 classes that represent various targets in urban environments. The results show that the increased spatial context of the multi-scan format improves the segmentation performance on the single-wavelength lidar dataset from 45.4 mIoU (a single scan) to 62.1 mIoU (24 consecutive scans). In the multispectral point cloud experiments we achieved a 71 % and 28 % relative increase in the segmentation mIoU (43.5 mIoU) as compared to the purely single-wavelength reference experiments, in which we achieved 25.4 mIoU (905 nm) and 34.1 mIoU (1550 nm). Our findings show that it is possible to semantically segment 2D line scanner data with good results by combining consecutive scans without the need for odometry information. The results also serve as motivation for developing multispectral mobile laser scanning systems that can be used in challenging urban surveys.

Neural-ABC: Neural Parametric Models for Articulated Body With Clothes.

In this paper, we introduce Neural-ABC, a novel parametric model based on neural implicit functions that can represent clothed human bodies with disentangled latent spaces for identity, clothing, shape, and pose. Traditional mesh-based representations struggle to represent articulated bodies with clothes due to the diversity of human body shapes and clothing styles, as well as the complexity of poses. Our proposed model provides a unified framework for parametric modeling, which can represent the identity, clothing, shape and pose of the clothed human body. Our proposed approach utilizes the power of neural implicit functions as the underlying representation and integrates well-designed structures to meet the necessary requirements. Specifically, we represent the underlying body as a signed distance function and clothing as an unsigned distance function, and they can be uniformly represented as unsigned distance fields. Different types of clothing do not require predefined topological structures or classifications, and can follow changes in the underlying body to fit the body. Additionally, we construct poses using a controllable articulated structure. The model is trained on both open and newly constructed datasets, and our decoupling strategy is carefully designed to ensure optimal performance. Our model excels at disentangling clothing and identity in different shape and poses while preserving the style of the clothing. We demonstrate that Neural-ABC fits new observations of different types of clothing. Compared to other state-of-the-art parametric models, Neural-ABC demonstrates powerful advantages in the reconstruction of clothed human bodies, as evidenced by fitting raw scans, depth maps and images. We show that the attributes of the fitted results can be further edited by adjusting their identities, clothing, shape and pose codes. The dataset and trained parametric model will be available at https://ustc3dv.github.io/NeuralABC/.

Skeleton Extraction for Articulated Objects With the Spherical Unwrapping Profiles.

Embedding unified skeletons into unregistered scans is fundamental to finding correspondences, depicting motions, and capturing underlying structures among the articulated objects in the same category. Some existing approaches rely on laborious registration to adapt a predefined LBS model to each input, while others require the input to be set to a canonical pose, e.g., T-pose or A-pose. However, their effectiveness is always influenced by the water-tightness, face topology, and vertex density of the input mesh. At the core of our approach lies a novel unwrapping method, named SUPPLE (Spherical UnwraPping ProfiLEs), which maps a surface into image planes independent of mesh topologies. Based on this lower-dimensional representation, a learning-based framework is further designed to localize and connect skeletal joints with fully convolutional architectures. Experiments demonstrate that our framework yields reliable skeleton extractions across a broad range of articulated categories, from raw scans to online CADs.

An Enhanced Multi-Sensor Simultaneous Localization and Mapping (SLAM) Framework with Coarse-to-Fine Loop Closure Detection Based on a Tightly Coupled Error State Iterative Kalman Filter

In order to attain precise and robust transformation estimation in simultaneous localization and mapping (SLAM) tasks, the integration of multiple sensors has demonstrated effectiveness and significant potential in robotics applications. Our work emerges as a rapid tightly coupled LIDAR-inertial-visual SLAM system, comprising three tightly coupled components: the LIO module, the VIO module, and the loop closure detection module. The LIO module directly constructs raw scanning point increments into a point cloud map for matching. The VIO component performs image alignment by aligning the observed points and the loop closure detection module imparts real-time cumulative error correction through factor graph optimization using the iSAM2 optimizer. The three components are integrated via an error state iterative Kalman filter (ESIKF). To alleviate computational efforts in loop closure detection, a coarse-to-fine point cloud matching approach is employed, leverging Quatro for deriving a priori state for keyframe point clouds and NanoGICP for detailed transformation computation. Experimental evaluations conducted on both open and private datasets substantiate the superior performance of the proposed method compared to similar approaches. The results indicate the adaptability of this method to various challenging situations.

Quantitative Amyloid‐PET in Real‐World Practice: Lessons from the Imaging Dementia—Evidence for Amyloid Scanning (IDEAS) study

AbstractBackgroundAmyloid‐PET quantification was developed in research settings using highly selected samples, harmonized acquisition protocols, and MRI‐based preprocessing. We aimed to analyze scans from IDEAS, a large‐scale, real‐world study of amyloid‐PET. Major challenges lie in scans being acquired on various PET/PET‐CT/PET‐MR scanners without standardization of acquisition or reconstruction protocols, and without MRI to preprocess data.Method18,295 Medicare beneficiaries with MCI or dementia underwent PET with [18F]florbetapir, [18F]florbetaben, or [18F]flutemetamol at 343 facilities. Sites shared 10,700 raw scans with their visual interpretation (positive or negative) by local radiologists/nuclear medicine physicians. Scans were processed using a PET‐only pipeline, which derives Centiloid based on a cortex‐to‐cerebellum tissue ratio in template space. Processed scans underwent standard visual quality control by 2 raters (Fig 1a‐b), and flagged scans were reviewed in a consensus meeting to discuss troubleshooting options. Injected dose and time between tracer injection and PET acquisition start were obtained from DICOM file header.Result9,963 scans (93%) were successfully pre‐processed with default parameters (Fig 1a) and 403 additional scans (4%) were included after manual intervention (Fig 1c). 339 scans (3%) were excluded because of technical artifacts or major bilateral lesions (Fig 1c). The final sample (n = 10,361) is described in Table 1, with example scans in Fig 1d. Centiloids showed the expected bimodal distribution (Fig 2a) and differentiated visually positive from negative scans (median = ‐2 versus 74 CL; area under the curve = 0.913, Fig 2b). PET acquisition generally followed tracer‐specific FDA recommendations for injected dose and acquisition time, but major variability was observed (Fig 2c). Centiloid values were independent from injected dose but were higher in patients scanned at later time windows for [18F]flutemetamol, resulting from known slow tracer kinetics (Fig 2c).ConclusionMRI‐free, amyloid‐PET quantification in a real‐world setting was feasible, with results conforming to expected distribution of Centiloid values and concordant with visual reads. However, greater standardization of PET acquisition than is currently required by FDA labels will be needed to reliably quantify amyloid PET in clinical practice (e.g., to monitor response to amyloid lowering therapy). The IDEAS dataset will be made available via the Global Alzheimer’s Association Interactive Network in 2023.

Applications of hybrid structure light three‐dimensional scanning in neurodegenerative disease autopsy brain banking

AbstractBackgroundPostmortem magnetic resonance imaging (MRI) and three‐dimensional (3D) printing has been demonstrated to be a feasible approach for the creation of tissue matrices to aid in postmortem brain dissection. We aimed to extend this approach using hybrid structure light technology (visible and infrared) 3D scanning to generate brain models as a digital resource and operational tool for human neurodegenerative disease brain banks.MethodA consecutive series of 56 brains accessioned into the Glenn Biggs Institute Brain Bank were imaged using an EinScan H handheld 3D scanner. Five brains were selected for generation of tissue matrices, with infratentorial structures removed using 3D object post‐processing software tools. Matrices were printed using a Raise3D Pro2 printer and 1.75‐mm polylactic acid premium filament.ResultEach scan was approximately an hour to complete, including capture and alignment of a lateral and a medial surface scan for each brain. Raw scan files (DAT and BRAP‐PRJ) averaged 5.47 GB and 3D object files were 12,991 KB for 3MF and 18,067 KB for OBJ file types. For cases with tissue matrices, average print time was 42 hours with an average filament consumption of 338 g. Brain cutting with the tissue matrices was performed in a standardized manner, with superb control for section thickness and planar orientation compared to brains cut without a matrix.Conclusion3D scanning of postmortem autopsied brains is a simple and useful strategy to generate models with utility in guided tissue sectioning and storage (practice), surface area volumetric analysis (research) and virtual reality/augmented reality neuroanatomy training (education) for Alzheimer’s disease and related dementias.

Multi-Slice Generation sMRI and fMRI for Autism Spectrum Disorder Diagnosis Using 3D-CNN and Vision Transformers.

Researchers have explored various potential indicators of ASD, including changes in brain structure and activity, genetics, and immune system abnormalities, but no definitive indicator has been found yet. Therefore, this study aims to investigate ASD indicators using two types of magnetic resonance images (MRI), structural (sMRI) and functional (fMRI), and to address the issue of limited data availability. Transfer learning is a valuable technique when working with limited data, as it utilizes knowledge gained from a pre-trained model in a domain with abundant data. This study proposed the use of four vision transformers namely ConvNeXT, MobileNet, Swin, and ViT using sMRI modalities. The study also investigated the use of a 3D-CNN model with sMRI and fMRI modalities. Our experiments involved different methods of generating data and extracting slices from raw 3D sMRI and 4D fMRI scans along the axial, coronal, and sagittal brain planes. To evaluate our methods, we utilized a standard neuroimaging dataset called NYU from the ABIDE repository to classify ASD subjects from typical control subjects. The performance of our models was evaluated against several baselines including studies that implemented VGG and ResNet transfer learning models. Our experimental results validate the effectiveness of the proposed multi-slice generation with the 3D-CNN and transfer learning methods as they achieved state-of-the-art results. In particular, results from 50-middle slices from the fMRI and 3D-CNN showed a profound promise in ASD classifiability as it obtained a maximum accuracy of 0.8710 and F1-score of 0.8261 when using the mean of 4D images across the axial, coronal, and sagittal. Additionally, the use of the whole slices in fMRI except the beginnings and the ends of brain views helped to reduce irrelevant information and showed good performance of 0.8387 accuracy and 0.7727 F1-score. Lastly, the transfer learning with the ConvNeXt model achieved results higher than other transformers when using 50-middle slices sMRI along the axial, coronal, and sagittal planes.

A novel post-processing strategy to improve the accuracy of complete-arch intraoral scanning for implants: an in vitro study

ObjectivesTo develop a new post-processing strategy that utilizes an auxiliary device to adjust intraoral scans and improve the accuracy of 3D models of complete-arch dental implants. Materials and methodsAn edentulous resin model with 6 dental implants was prepared. An auxiliary device, consisting of an opaque base and artificial landmarks, was fabricated and mounted onto the resin model. Twenty intraoral scans (raw scans) were taken using this setup. A new post-processing strategy was proposed to adjust the raw scans using reverse engineering software (verified group). Additionally, ten conventional gypsum casts were duplicated and digitized using a laboratory scanner. The linear and angular trueness and precision of the models were evaluated and compared. The effect of the proposed strategy on the accuracy of complete-arch intraoral scans was analyzed using one-way ANOVA. ResultsThe linear trueness (29.7 µm) and precision (24.8 µm) of the verified group were significantly better than the raw scans (46.6 µm, 44.7 µm) and conventional casts (51.3 µm, 36.5 µm), particularly in cross-arch sites. However, the angular trueness (0.114°) and precision (0.085°) of the conventional casts were significantly better than both the verified models (0.298°, 0.168°) and the raw scans (0.288°, 0.202°). ConclusionsThe novel post-processing strategy is effective in enhancing the linear accuracy of complete-arch implant IO scans, especially in cross-arch sites. However, further improvement is needed to eliminate the angular deviations. Clinical significanceErrors generated from intraoral scanning in complete edentulous arches exceed the clinical threshold. The elimination of stitching errors in the raw scans particularly in the cross-arch sites, through the proposed post-processing strategy would enhance the accuracy of complete-arch implant prostheses.

A Comparative Study of Semantic Segmentation using Deep Neural Networks in a GNSS-denied Underground Parking Lot

Abstract. Deep neural networks (DNNs) in intelligent point cloud processing have achieved remarkable progress in recent years. Most existing methods and models were adopted on either outdoor or indoor scenes while very few previous studies were conducted in GNSS-denied environments. In this paper, we carried out a comparative study in semantic segmentation outputs using different DNNs in an underground parking lot dataset. Manually labeled indoor point cloud data were trained and tested using 7 different DNNs (e.g. PointNet, KPConv, FPConv, BAAF-Net, etc.). Our experiments demonstrated how well different DNNs perform in GNSS-denied environments with performance assessments in mIoU, Mean Accuracy (mAcc), Overall Accuracy (OA), as well as visualization outputs. The main contribution of this comparative study is to compare state-of-the-art DNN algorithms’ performance in semantic segmentation directly on the raw indoor mobile laser scanning (iMLS) data from a GNSS-denied underground parking lot and evaluate the effectiveness and potentials of different DNNs in underground 3D taskings. Draw upon that, which current algorithms are optimal and how future work in GNSS-denied environments can be inspired and implemented would be discussed.

Deep learning-based automated lesion segmentation on mouse stroke magnetic resonance images

Magnetic resonance imaging (MRI) is widely used for ischemic stroke lesion detection in mice. A challenge is that lesion segmentation often relies on manual tracing by trained experts, which is labor-intensive, time-consuming, and prone to inter- and intra-rater variability. Here, we present a fully automated ischemic stroke lesion segmentation method for mouse T2-weighted MRI data. As an end-to-end deep learning approach, the automated lesion segmentation requires very little preprocessing and works directly on the raw MRI scans. We randomly split a large dataset of 382 MRI scans into a subset (n = 293) to train the automated lesion segmentation and a subset (n = 89) to evaluate its performance. We compared Dice coefficients and accuracy of lesion volume against manual segmentation, as well as its performance on an independent dataset from an open repository with different imaging characteristics. The automated lesion segmentation produced segmentation masks with a smooth, compact, and realistic appearance that are in high agreement with manual segmentation. We report dice scores higher than the agreement between two human raters reported in previous studies, highlighting the ability to remove individual human bias and standardize the process across research studies and centers.

Comparison of Anatomical and Diffusion MRI for detecting Parkinson's Disease using Deep Convolutional Neural Network.

Parkinson's disease (PD) is a progressive neurodegenerative disease that affects over 10 million people worldwide. Brain atrophy and microstructural abnormalities tend to be more subtle in PD than in other age-related conditions such as Alzheimer's disease, so there is interest in how well machine learning methods can detect PD in radiological scans. Deep learning models based on convolutional neural networks (CNNs) can automatically distil diagnostically useful features from raw MRI scans, but most CNN-based deep learning models have only been tested on T1-weighted brain MRI. Here we examine the added value of diffusion-weighted MRI (dMRI) - a variant of MRI, sensitive to microstructural tissue properties - as an additional input in CNN-based models for PD classification. Our evaluations used data from 3 separate cohorts - from Chang Gung University, the University of Pennsylvania, and the PPMI dataset. We trained CNNs on various combinations of these cohorts to find the best predictive model. Although tests on more diverse data are warranted, deep-learned models from dMRI show promise for PD classification.Clinical Relevance- This study supports the use of diffusion-weighted images as an alternative to anatomical images for AI-based detection of Parkinson's disease.

Deep survival modeling of longitudinal retinal OCT volumes for predicting the onset of atrophy in patients with intermediate AMD.

In patients with age-related macular degeneration (AMD), the risk of progression to late stages is highly heterogeneous, and the prognostic imaging biomarkers remain unclear. We propose a deep survival model to predict the progression towards the late atrophic stage of AMD. The model combines the advantages of survival modelling, accounting for time-to-event and censoring, and the advantages of deep learning, generating prediction from raw 3D OCT scans, without the need for extracting a predefined set of quantitative biomarkers. We demonstrate, in an extensive set of evaluations, based on two large longitudinal datasets with 231 eyes from 121 patients for internal evaluation, and 280 eyes from 140 patients for the external evaluation, that this model improves the risk estimation performance over standard deep learning classification models.

A Lightweight and Detector-Free 3D Single Object Tracker on Point Clouds

Recent works on 3D single object tracking treat the task as a target-specific 3D detection task, where an off-the-shelf 3D detector is commonly employed for the tracking. However, it is non-trivial to perform accurate target-specific detection since the point cloud of objects in raw LiDAR scans is usually sparse and incomplete. In this paper, we address this issue by explicitly leveraging temporal motion cues and propose DMT, a Detector-free Motion-prediction-based 3D Tracking network that completely removes the usage of complicated 3D detectors and is lighter, faster, and more accurate than previous trackers. Specifically, the motion prediction module is first introduced to estimate a potential target center of the current frame in a point-cloud-free manner. Then, an explicit voting module is proposed to directly regress the 3D box from the estimated target center. Extensive experiments on KITTI and NuScenes datasets demonstrate that our DMT can still achieve better performance ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\sim$</tex-math> </inline-formula> 10% improvement over the NuScenes dataset) and a faster tracking speed (i.e., 72 FPS) than state-of-the-art approaches without applying any complicated 3D detectors. Our code is released at https://github.com/jimmy-dq/DMT.

Automated and data-driven plate computation for presurgical cleft lip and palate treatment

PurposePresurgical orthopedic plates are widely used for the treatment of cleft lip and palate, which is the most common craniofacial birth defect. For the traditional plate fabrication, an impression is taken under airway-endangering conditions, which recent digital alternatives overcome via intraoral scanners. However, these alternatives demand proficiency in 3D modeling software in addition to the generally required clinical knowledge of plate design.MethodsWe address these limitations with a data-driven and fully automated digital pipeline, endowed with a graphical user interface. The pipeline adopts a deep learning model to landmark raw intraoral scans of arbitrary mesh topology and orientation, which guides the nonrigid surface registration subsequently employed to segment the scans. The plates that are individually fit to these segmented scans are 3D-printable and offer optional customization.ResultsWith the distance to the alveolar ridges closely centered around the targeted 0.1 mm, our pipeline computes tightly fitting plates in less than 3 min. The plates were approved in 12 out of 12 cases by two cleft care professionals in a printed-model-based evaluation. Moreover, since the pipeline was implemented in clinical routine in two hospitals, 19 patients have been undergoing treatment utilizing our automated designs.ConclusionThe results demonstrate that our automated pipeline meets the high precision requirements of the medical setting employed in cleft lip and palate care while substantially reducing the design time and required clinical expertise, which could facilitate access to this presurgical treatment, especially in low-income countries.

DeepMiCa: Automatic segmentation and classification of breast MIcroCAlcifications from mammograms

Background and objectiveBreast cancer is the world’s most prevalent form of cancer. The survival rates have increased in the last years mainly due to factors such as screening programs for early detection, new insights on the disease mechanisms as well as personalised treatments. Microcalcifications are the only first detectable sign of breast cancer and diagnosis timing is strongly related to the chances of survival. Nevertheless microcalcifications detection and classification as benign or malignant lesions is still a challenging clinical task and their malignancy can only be proven after a biopsy procedure. We propose DeepMiCa, a fully automated and visually explainable deep-learning based pipeline for the analysis of raw mammograms with microcalcifications. Our aim is to propose a reliable decision support system able to guide the diagnosis and help the clinicians to better inspect borderline difficult cases. MethodsDeepMiCa is composed by three main steps: (1) Preprocessing of the raw scans (2) Automatic patch-based Semantic Segmentation using a UNet based network with a custom loss function appositely designed to deal with extremely small lesions (3) Classification of the detected lesions with a deep transfer-learning approach. Finally, state-of-the-art explainable AI methods are used to produce maps for a visual interpretation of the classification results. Each step of DeepMiCa is designed to address the main limitations of the previous proposed works resulting in a novel automated and accurate pipeline easily customisable to meet radiologists’ needs. ResultsThe proposed segmentation and classification algorithms achieve an area under the ROC curve of 0.95 and 0.89 respectively. Compared to previously proposed works, this method does not require high performance computational resources and provides a visual explanation of the final classification results. ConclusionTo conclude, we designed a novel fully automated pipeline for detection and classification of breast microcalcifications. We believe that the proposed system has the potential to provide a second opinion in the diagnosis process giving the clinicians the opportunity to quickly visualise and inspect relevant imaging characteristics. In the clinical practice the proposed decision support system could help reduce the rate of misclassified lesions and consequently the number of unnecessary biopsies.

OzNet: A New Deep Learning Approach for Automated Classification of COVID-19 Computed Tomography Scans.

Coronavirus disease 2019 (COVID-19) is spreading rapidly around the world. Therefore, the classification of computed tomography (CT) scans alleviates the workload of experts, whose workload increased considerably during the pandemic. Convolutional neural network (CNN) architectures are successful for the classification of medical images. In this study, we have developed a new deep CNN architecture called OzNet. Moreover, we have compared it with pretrained architectures namely AlexNet, DenseNet201, GoogleNet, NASNetMobile, ResNet-50, SqueezeNet, and VGG-16. In addition, we have compared the classification success of three preprocessing methods with raw CT scans. We have not only classified the raw CT scans, but also have performed the classification with three different preprocessing methods, which are discrete wavelet transform (DWT), intensity adjustment, and gray to color red, green, blue image conversion on the data sets. Furthermore, it is known that the architecture's performance increases with the use of DWT preprocessing method rather than using the raw data set. The results are extremely promising with the CNN algorithms using the COVID-19 CT scans processed with the DWT. The proposed DWT-OzNet has achieved a high classification performance of more than 98.8% for each calculated metric.

CNN-Based Fault Detection of Scan Matching for Accurate SLAM in Dynamic Environments.

This paper proposes a method for CNN-based fault detection of the scan-matching algorithm for accurate SLAM in dynamic environments. When there are dynamic objects in an environment, the environment that is detected by a LiDAR sensor changes. Thus, the scan matching of laser scans is likely to fail. Therefore, a more robust scan-matching algorithm to overcome the faults of scan matching is needed for 2D SLAM. The proposed method first receives raw scan data in an unknown environment and executes ICP (Iterative Closest Points) scan matching of laser scans from a 2D LiDAR. Then, the matched scans are converted into images, which are fed into a CNN model for its training to detect the faults of scan matching. Finally, the trained model detects the faults when new scan data are provided. The training and evaluation are performed in various dynamic environments, taking real-world scenarios into account. Experimental results showed that the proposed method accurately detects the faults of scan matching in every experimental environment.