False Positive Reduction Research Articles

A Transfer Learning-Based Approach to Detect Cerebral Microbleeds

Cerebral microbleeds are small chronic vascular diseases that occur because of irregularities in the cerebrum vessels. Individuals and elderly people with brain injury and dementia can have small microbleeds in their brains. A recent study has shown that cerebral microbleeds could be remarkably risky in terms of life and can be riskier for patients with dementia. In this study, we proposed an efficient approach to automatically identify microbleeds by reducing the false positives in openly available susceptibility-weighted imaging (SWI) data samples. The proposed structure comprises two different pre-trained convolutional models with four stages. These stages include (i) skull removal and augmentation, (ii) making clusters of data samples using the k-mean classifier, (iii) reduction of false positives for efficient performance, and (iv) transfer-learning classification. The proposed technique was assessed using the SWI dataset available for 20 subjects. For our findings, we attained an accuracy of 97.26% with a 1.8% false-positive rate using data augmentation on the AlexNet transfer learning model and a 1.1% false-positive rate with 97.89% accuracy for the ResNet 50 model with data augmentation approaches. The results show that our models outperformed the existing approach for the detection of microbleeds.

Kidney tumor segmentation from computed tomography images using DeepLabv3+ 2.5D model

Kidney cancer is a public health problem that affects thousands of people worldwide. Accurate kidney tumor segmentation is an important task that helps doctors to reduce the time to interpret images, diagnose, and plan the treatment of the diseases. However, because of the heterogeneity of tumors, manually segmenting is a time-consuming process and subject to variability among specialists. In the last years, convolutional neural networks (CNNs) have gained popularity for automatic medical image segmentation. Three-dimensional (3D) networks have a high segmentation capacity, but they are complex and have high computational costs. Thus, two-dimensional networks are the most used owing to the relatively low memory consumption, but they do not exploit 3D features. Therefore, in the present study, a 2.5D network that balances memory consumption and model complexity is proposed to assist doctors specialized in kidney tumor diagnosis in computed tomography (CT). There are three major steps in the proposed method: (1) preprocessing of the KiTS19 dataset, (2) initial segmentation using the DeepLabv3+ 2.5D model with DPN-131 encoder, and (3) reduction of false positives using image processing. The performance of the proposed method was assessed in 210 CTs from the available KiTS19 dataset. In the best scenario, the proposed method presented accuracy of 99.71%, sensitivity of 84.24%, specificity of 99.82%, Dice coefficient of 85.17%, Jaccard index of 75.62%, Hausdorff distance of 18.39 mm, and average surface distance of 3.36 mm. Thus, the 2.5D neural network approach is capable of efficiently solving the problem of segmentation of kidney tumors in CT images, presenting results comparable to those of high-performance 3D neural networks.

High-Capacity Real-Time Face Retrieval Recognition Algorithm Based on Task Scheduling Model for the Treatment Area of Hospital

This paper presents an in-depth study of face detection, face feature extraction, and face classification from three important components of a high-capacity face recognition system for the treatment area of hospital and a study of a high-capacity real-time face retrieval and recognition algorithm for the treatment area of hospital based on a task scheduling model. Considering the real-time nature of our system, our face feature extraction network is modeled by DeepID, and the network is slightly improved by introducing a central loss verification signal to train a DeepID-like network model using central loss and use it to extract face features. To further investigate and optimize the schedulability analysis problem of the directed graph real-time task model, this paper proposes a rigorous and approximate response time analysis method for the directed graph real-time task model with an arbitrary time frame. Based on the theoretical results of the greatly additive algebra, it is shown that the coherent qualifying function is linearly periodic, i.e., the function can be represented by a finite nonperiodic part and an infinitely repeated periodic part, thus calculating the coherent qualifying function independent of the magnitude of the interval time. The algorithm for high-capacity real-time face retrieval and recognition in the treatment area of hospital based on the task scheduling model is further investigated, and a face database is established by using the PCA dimensionality reduction technique. Based on the internal architecture of the processor, image preprocessing and IP core packaging are implemented, and the hardware engineering of the high-capacity real-time face recognition system for hospital visits is built using the IP-based design concept. The performance tests of the face detection model and feature extraction network show that the face detection model has a significant reduction in false-positive rate, better fitting of border regression, and improved time performance. The face feature extraction network has no overfitting, and the features are highly discriminative with small feature extraction time consumption. The high-capacity real-time face recognition system for the treatment area of hospital combined with the optimized directed graph task scheduling model can approach 25 fps, which meets the real-time requirements, and the face recognition rate surpasses that of real people. It realizes the intelligence, self-help, and autonomy of medical services and satisfies the medical needs of users in all aspects.

Money laundering transaction detection with classification tree models

The detection of money laundering is a very important problem, especially in the financial sector. We propose a mathematical specification of the problem in terms of a classification tree model that ”automates” expert based manual decisions. We operationally validate the model on a concrete application that originates from a large Italian bank. The application of the model to the data shows a good predictive accuracy and, even more importantly, the reduction of false positives, with respect to the ”manual” expert based activity. From an interpretational viewpoint, while some drivers of suspicious laundering activity are in line with the daily business practices of the bank’s anti money laundering operations, some others are new discoveries.

ALA-Net: Adaptive Lesion-Aware Attention Network for 3D Colorectal Tumor Segmentation.

Accurate and reliable segmentation of colorectal tumors and surrounding colorectal tissues on 3D magnetic resonance images has critical importance in preoperative prediction, staging, and radiotherapy. Previous works simply combine multilevel features without aggregating representative semantic information and without compensating for the loss of spatial information caused by down-sampling. Therefore, they are vulnerable to noise from complex backgrounds and suffer from misclassification and target incompleteness-related failures. In this paper, we address these limitations with a novel adaptive lesion-aware attention network (ALA-Net) which explicitly integrates useful contextual information with spatial details and captures richer feature dependencies based on 3D attention mechanisms. The model comprises two parallel encoding paths. One of these is designed to explore global contextual features and enlarge the receptive field using a recurrent strategy. The other captures sharper object boundaries and the details of small objects that are lost in repeated down-sampling layers. Our lesion-aware attention module adaptively captures long-range semantic dependencies and highlights the most discriminative features, improving semantic consistency and completeness. Furthermore, we introduce a prediction aggregation module to combine multiscale feature maps and to further filter out irrelevant information for precise voxel-wise prediction. Experimental results show that ALA-Net outperforms state-of-the-art methods and inherently generalizes well to other 3D medical images segmentation tasks, providing multiple benefits in terms of target completeness, reduction of false positives, and accurate detection of ambiguous lesion regions.

Locating illicit discharges in storm sewers in urban areas using multi-parameter source tracking: Field validation of a toolbox composite index to prioritize high risk areas

In urban areas served by separate sewerage systems, illicit connections to the storm drain system from residences or commercial establishments are frequent whether these misconnections were made accidentally or deliberately. As a result, untreated and contaminated wastewater enters into storm sewers leading to pollution of receiving waters and non-compliance with water quality standards. Typical procedures for detecting illicit connections to the storm sewer system are time consuming and expensive, especially in a highly urbanised area. In this study, we investigated the use of human wastewater micropollutants WWMPs (caffeine, theophylline, and carbamazepine) and advanced DNA molecular markers (human specific Bacteroides HF183 and mitochondrial DNA) as anthropogenic tracers in order to assist identifying potential cross connections.Water samples from storm outfalls and storm sewer pipes in three urban subcatchments were collected in dry weather from 2013 to 2018. All samples contained various concentrations of these markers especially HF183, caffeine and theophylline, suggesting that the storm pipe system studied is widely contaminated by sanitary sewers. None of the traditional indicators or markers tested is sufficient alone to determine the origin of fecal pollution. In a highly urbanised area, the combination of at least three specific human markers was needed in order to locate the residential section with likely misconnections. The human specific Bacteroides HF183, and theophylline appeared to be the most effective markers (along with E. coli) of crossconnections, whereas carbamazepine can provide an indication of contamination through sanitary sewer exfiltration. A composite sewer cross-connection index was developed, and eight misconnected houses were identified and corrected. The index approach enables the reduction of false positives that could lead to expensive interventions to identify cross-connected households. The results show the multiparameter source tracking toolbox as an effective method to identify sewer cross connections for sustainable storm water management.

Automated Detection of Gastric Cancer by Retrospective Endoscopic Image Dataset Using U-Net R-CNN

Upper gastrointestinal endoscopy is widely performed to detect early gastric cancers. As an automated detection method for early gastric cancer from endoscopic images, a method involving an object detection model, which is a deep learning technique, was proposed. However, there were challenges regarding the reduction in false positives in the detected results. In this study, we proposed a novel object detection model, U-Net R-CNN, based on a semantic segmentation technique that extracts target objects by performing a local analysis of the images. U-Net was introduced as a semantic segmentation method to detect early candidates for gastric cancer. These candidates were classified as gastric cancer cases or false positives based on box classification using a convolutional neural network. In the experiments, the detection performance was evaluated via the 5-fold cross-validation method using 1208 images of healthy subjects and 533 images of gastric cancer patients. When DenseNet169 was used as the convolutional neural network for box classification, the detection sensitivity and the number of false positives evaluated on a lesion basis were 98% and 0.01 per image, respectively, which improved the detection performance compared to the previous method. These results indicate that the proposed method will be useful for the automated detection of early gastric cancer from endoscopic images.

Abstract 1122‐000235: Artificial Intelligence Algorithms for Hemorrhage Detection in CTs and MRI Scans: A Systematic Review

Introduction : Stroke is a leading cause of morbidity and mortality worldwide, with hemorrhagic strokes accounting for 10–20% of all strokes. Patients presenting with intracerebral hemorrhage (ICH) often face higher rates of mortality and poorer prognosis than those with other stroke types. As ICH treatment relies on in‐hospital neuroimaging findings, one potential barrier in the effective management of ICH includes increased time to ICH detection and treatment, particularly due to delays in imaging interpretation in busy hospitals and emergency departments. Artificial Intelligence (AI) driven software has recently been developed and become commercially available for the detection of Intracranial Hemorrhage (ICH) and Chronic Cerebral Microbleeds (CMBs). Such adjunct tools may enhance patient care by decreasing time to treatment and diagnosis by helping to adjudicate diagnoses in difficult cases. This systematic review aims to describe the current literature surrounding all currently existing AI algorithms for ICH detection with either non‐contrast computed tomography (CT) scans or CMBs detection with magnetic resonance imaging (MRI). Methods : Following PRISMA guidelines, MEDLINE and EMBASE were searched for studies published through March 1st, 2021, and all studies investigating AI algorithms for hemorrhage detection in non‐contrast CT scans or CMBs detection on MRI scans were eligible for inclusion. Any studies focusing on AI for hemorrhage segmentation only, including studies that enrolled patients with hemorrhages only as their study group, were excluded. Extracted data included development methods, training, validation and testing datasets, and accuracy metrics for each algorithm, when available. Meta‐analysis was not conducted due to heterogeneity in reported accuracy metrics and highly variant algorithmic development. The completed protocol is available for review in the PROSPERO registry. Results : After the removal of duplicates, a total of 609 studies were identified and screened. After an initial screening and full text review, 40 studies were included in this review. Of these, 18 tested a 2‐Dimensional (2D) convolutional neural network (CNN) AI algorithm, 3 used a purley 3‐Dimension (3D) CNN, and 2 utilized a hybrid 2D‐3D CNN. Of note, one software was able to identify ICH in the setting of ischemic stroke using MRI scans. Included papers noted the following challenges when developing these AI algorithms: extensive time required to create suitable datasets, the volumetric nature of the imaging exams, fine tuning the system, and focusing on the reduction of false positives. Diagnostic accuracy data was available for 21 of these studies, which reported a mean accuracy of 94.37% and a mean AUC of 0.958. Conclusions : As reported in this study, many AI‐driven software tools have been developed over the last 5 years. These tools have high diagnostic accuracy on average and have the potential to contribute to the diagnosis of ICH or CMBs with expert‐level accuracy. With time to treatment often dependent on time to diagnosis, this AI software may increase both the speed and accuracy of adjudicating diagnoses. Although there have been several obstacles faced by the developers of these algorithms, AI‐driven software is an important frontier for the future of clinical medicine.

Le recours aux dossiers des patients est-il nécessaire pour l’évaluation du risque de récidive sexuelle, violente et générale ?

IntroductionSexual, violent and general recidivism risk scales are widely used in a number of countries. Their psychometric qualities are generally considered to be good. However, in practice they may vary in the quality of prediction of risk of sexual, violent and general recidivism, in particular because of the sources of the information collected. In France, the medical records of incarcerated patients are kept by health-care professionals. Although regulated, the content and quality of these records vary widely from one patient to another. The criminal justice system holds the criminal records of convicted and imprisoned persons. There is no set list of documents contained in these records. For caregivers and researchers, access to criminal records is difficult because of the confidentiality to which legal professionals are subject. The aim of our study was to investigate whether using medical files in addition to structured interviews can improve the assessment and management of the risks of sexual, violent and general recidivism. Material and MethodA total of 128 perpetrators of violence were assessed using three scales of risk of sexual, violent and general recidivism. Scores for the items of the scales were compared between (a) those that were based on medical records and an interview, and (b) those based only on an interview. Results and discussionFirst, differences in scores between the two groups (assessed through interview only, and assessed through interview and use of medical records) were observed on the RSVP, HCR-20 and LS/CMI scales. Secondly, most of the results indicate that the overall level of risk was perceived as lower when medical records were used, which would, indirectly, lead to a reduction in false positives when evaluating perpetrators of sexual violence. Thirdly, the point-by-point analysis shows that the use of information contained in the medical records reduces the weight of present and future factors (e.g. the physical and psychological stress of recent events), increases the weight of past factors (e.g. history of sexual violence), and can increase the weight of certain factors that can lead to more negative emotions in the assessor (e.g. deviance). These results can be explained by (i) the emotional functioning of the persons assessed (particularly defensive processes or memory difficulties), (ii) the attitude of the aggressor (particularly the presence of emotional and cognitive biases), (iii) the nature of the information (particularly “hot” cognitions or those leading to greater social desirability). The limitations of the study concern the relatively small number of participants, the environment in which the file was transmitted and the very heterogeneous and sometimes relatively incomplete composition of the files. ConclusionThe use of information contained in medical files impacts the results of recidivism risk scales and restores a balance to the factors. In France, risk scales are currently being introduced, although their use is still limited in the health field. However, ethical use of these scales raises the issue of the homogenisation of the content of medical records and access to criminal records in order to enable future research to confirm whether the use of information provided in medical and criminal records can improve the quality of assessment and treatment of offenders.

A Deep Learning Based Brain Metastases Segmentation Web Platform for Stereotactic Radiosurgery

Stereotactic radiosurgery (SRS) enables the treatment of multiple (> 4) brain metastases (mBMs) patients with reduced neurotoxicity. BMs delineation serves as an essential part in the SRS management involving treatment planning and follow-up. However, manual contouring is highly time-consuming, especially in patients with mBMs. In this study, we propose a deep learning-based BMs segmentation web platform including BMs labeling and database management to improve the efficiency of SRS workflow.The platform is designed with a web client and a back-end server. The web client enables user interactions and data visualization, while the back-end server embeds the core algorithms and maintains a database. All the embedded deep learning models were previously-trained using 271 T1c MRI data with 1058 BMs in total. Once the user imports new MRI data from the client, the server will execute tasks including: 1) CNN-based BMs auto-segmentation 2) Siamese network based false-positive segmentation reduction 3) Affine registration-based BMs labeling, and then send the results back to the web client. The web client displays the received segmentation results, and allows the users to conduct further modifications and export as DICOM RTSTRUCT files. In addition, the server will update the database with the new data. Users can access this database from the web client to retrieve any patient data including image, segmentation and other metadata.The entire process takes about 4 to 5 minutes to complete per new data. The platform can segment mBMs with initial averaged sensitivity of 0.96 and specificity of 0.59. Then after the false-positive reduction, the averaged sensitivity and specificity currently are 0.86 and 0.89. Due to the flexibility of false-positive reduction algorithm, we can balance the sensitivity/specificity as clinical need. The segmentation accuracy is evaluated by the mean (std) Hausdorff distance as 2.75 (0.57) mm, the mean (std) center of mass shift as 1.41 (0.32) mm, and the mean (std) of surface-to-surface distance as 1.01 (0.28) mm, with the BMs labeling accuracy as 100%.The deep learning BMs segmentation web platform can accurately segment and label mBMs, and maintain a database for improved data management. It can facilitate the efficiency of the SRS workflow including planning and treatment follow-up, and can be a promising tool for the BMs management.

Automatic quantification of white matter hyperintensities on T2-weighted fluid attenuated inversion recovery magnetic resonance imaging

White matter hyperintensities (WMH) are areas of increased signal visualized on T2-weighted fluid attenuated inversion recovery (FLAIR) brain magnetic resonance imaging (MRI) sequences. They are typically attributed to small vessel cerebrovascular disease in the context of aging. Among older adults, WMH are associated with risk of cognitive decline and dementia, stroke, and various other health outcomes. There has been increasing interest in incorporating quantitative WMH measurement as outcomes in clinical trials, observational research, and clinical settings. Here, we present a novel, fully automated, unsupervised detection algorithm for WMH segmentation and quantification. The algorithm uses a robust preprocessing pipeline, including brain extraction and a sample-specific mask that incorporates spatial information for automatic false positive reduction, and a half Gaussian mixture model (HGMM). The method was evaluated in 24 participants with varying degrees of WMH (4.9–78.6 cm3) from a community-based study of aging and dementia with dice coefficient, sensitivity, specificity, correlation, and bias relative to the ground truth manual segmentation approach performed by two expert raters. Results were compared with those derived from commonly used available WMH segmentation packages, including SPM lesion probability algorithm (LPA), SPM lesion growing algorithm (LGA), and Brain Intensity AbNormality Classification Algorithm (BIANCA). The HGMM algorithm derived WMH values that had a dice score of 0.87, sensitivity of 0.89, and specificity of 0.99 compared to ground truth. White matter hyperintensity volumes derived with HGMM were strongly correlated with ground truth values (r = 0.97, p = 3.9e-16), with no observable bias (−1.1 [−2.6, 0.44], p-value = 0.16). Our novel algorithm uniquely uses a robust preprocessing pipeline and a half-Gaussian mixture model to segment WMH with high agreement with ground truth for large scale studies of brain aging.

An efficient multi-path 3D convolutional neural network for false-positive reduction of pulmonary nodule detection.

Considering that false-positive and true pulmonary nodules are highly similar in shapes and sizes between lung computed tomography scans, we develop and evaluate a false-positive nodules reduction method applied to the computer-aided diagnosis system. To improve the pulmonary nodule diagnosis quality, a 3D convolutional neural networks (CNN) model is constructed to effectively extract spatial information of candidate nodule features through the hierarchical architecture. Furthermore, three paths corresponding to three receptive field sizes are adopted and concatenated in the network model, so that the feature information is fully extracted and fused to actively adapting to the changes in shapes, sizes, and contextual information between pulmonary nodules. In this way, the false-positive reduction is well implemented in pulmonary nodule detection. Multi-path 3D CNN is performed on LUNA16 dataset, which achieves an average competitive performance metric score of 0.881, and excellent sensitivity of 0.952 and 0.962 occurs to 4, 8 FP/Scans. By constructing a multi-path 3D CNN to fully extract candidate target features, it accurately identifies pulmonary nodules with different sizes, shapes, and background information. In addition, the proposed general framework is also suitable for similar 3D medical image classification tasks.

An Iterative Algorithm for Semisupervised Classification of Hotspots on Bone Scintigraphies of Patients with Prostate Cancer.

Prostate cancer (PCa) is the second most diagnosed cancer in men. Patients with PCa often develop metastases, with more than 80% of this metastases occurring in bone. The most common imaging technique used for screening, diagnosis and follow-up of disease evolution is bone scintigraphy, due to its high sensitivity and widespread availability at nuclear medicine facilities. To date, the assessment of bone scans relies solely on the interpretation of an expert physician who visually assesses the scan. Besides this being a time consuming task, it is also subjective, as there is no absolute criteria neither to identify bone metastases neither to quantify them by a straightforward and universally accepted procedure. In this paper, a new algorithm for the false positives reduction of automatically detected hotspots in bone scintigraphy images is proposed. The motivation relies in the difficulty of building a fully annotated database. In this way, our algorithm is a semisupervised method that works in an iterative way. The ultimate goal is to provide the physician with a fast, precise and reliable tool to quantify bone scans and evaluate disease progression and response to treatment. The algorithm is tested in a set of bone scans manually labeled according to the patient’s medical record. The achieved classification sensitivity, specificity and false negative rate were 63%, 58% and 37%, respectively. Comparison with other state-of-the-art classification algorithms shows superiority of the proposed method.

Tips for a reduction of false positives in manual RT-PCR diagnostics of SARS-CoV-2

RT-PCR is the standard gold technique for testing the presence of RNA of the coronavirus causing Severe Acute Respiratory Syndrome (SARS-CoV-2) due to its high specificity and sensitivity. Despite its general use and reliability, no lab in the world is immune to the generation of false positives. These errors cause a loss of confidence in the technique's power and damage the image of laboratories. More importantly, they can take a toll on tested individuals and have economic, psychological, and health-associated effects. Most false positives are caused during a manual operation inside the laboratory. However, not much has been published about the errors associated with particular laboratory techniques used to detect the virus since the beginning of the actual pandemic. This work precisely reflects on events that occur during manual RT-PCR diagnostics in a COVID-19 laboratory, providing tips for reducing false-positive results.

Automatic breast mass detection in mammograms using density of wavelet coefficients and a patch-based CNN.

For the purpose of accurate and efficient mass detection in full-field digital mammograms, weproposeamethodfor automated mass detection that consists oftwostages: suspicious region localization and false-positive (FP) reduction, by classifying these regions into mass and non-mass regions (normal tissues). In the first stage, the density of the wavelet coefficients based on Quincunx Lifting Scheme (DWC-QLS) is used to find suspicious regions (regions of interest, ROIs) in full mammograms. In the second stage, a patch-based CNN classifier is developed as an FP reduction to classify the suspicious regions. The main aim of this stage is to reduce the false-positive suspicious regions while keeping the true-positive suspicious regions. To further improve the performance of the FP reduction, the effectiveness of different transfer learning strategies is further explored and the best fine-tuning strategy in training InceptionV3 model is determined experimentally. The experimental results show that the proposed method can achieve an overall performance of 0.98 TPR @1.43 FPI on the INbreast database. In addition, the suggested segmentation method detects the mass location with 100% sensitivity and average of 5.4 false positives per image. Based on the obtained results,the introduced method was able to successfully detectand classify suspicious regions in digital mammograms and provide better TPR and FPI results in comparison with other state-of-the-art method.

Lung Cancer Detection using CT SCAN Images

Detection of pulmonary nodules has a crucial effect on the diagnosis of lung cancer, but the detection is a nontrivial task, not only because the appearance of pulmonary nodules varies in a wide range, but also because nodule densities have low contrast against adjacent vessel segments and other lung tissues. Computed tomography (CT) has been shown as the most popular imaging modality for nodule detection, because it has the ability to provide reliable image textures for the detection of small nodules. The development of lung nodule CADe systems using CT imaging modality has made good progress over the past decade. Generally, such CADe systems consist of three stages: 1) image pre-processing, 2) initial nodule candidates (INCs) identification, and 3) false positive (FP) reduction of the INCs with preservation of the true positives (TPs). In the pre-processing stage, the system aims to largely reduce the search space to the lungs, where a segmentation of the lungs from the entire chest volume is usually required. Because of the high image contrast between lung fields and the surrounding body tissue, image intensity-based simple thresholding is effective, and is currently the most commonly used technique for lung segmentation. This paper proposes an adaptive solution to mitigate the difficulty of thresholding-based method in lung segmentation. Sufficient detection power for nodule candidates is inevitably accompanied by many (obvious) FPs. A rule-based filtering operation is often employed to cheaply and drastically reduce the number of obvious FPs, so that their influence on the computationally more expensive learning process can be eliminated. In general, FP reduction using machine learning has been extensively studied in the literature. Compared with unsupervised learning that aims to find hidden structures in unlabelled data, supervised learning, which aims to infer a function from labelled training data, is more frequently used to design a CADe system. Compared with the existing approaches, the morphology based lung cancer detection can be an alternative with either comparable detection performance and less computational cost, or comparable cost and better detection performance.

Automatic Handgun Detection with Deep Learning in Video Surveillance Images

There is a great need to implement preventive mechanisms against shootings and terrorist acts in public spaces with a large influx of people. While surveillance cameras have become common, the need for monitoring 24/7 and real-time response requires automatic detection methods. This paper presents a study based on three convolutional neural network (CNN) models applied to the automatic detection of handguns in video surveillance images. It aims to investigate the reduction of false positives by including pose information associated with the way the handguns are held in the images belonging to the training dataset. The results highlighted the best average precision (96.36%) and recall (97.23%) obtained by RetinaNet fine-tuned with the unfrozen ResNet-50 backbone and the best precision (96.23%) and F1 score values (93.36%) obtained by YOLOv3 when it was trained on the dataset including pose information. This last architecture was the only one that showed a consistent improvement—around 2%—when pose information was expressly considered during training.

End-to-end prostate cancer detection in bpMRI via 3D CNNs: Effects of attention mechanisms, clinical priori and decoupled false positive reduction

We present a multi-stage 3D computer-aided detection and diagnosis (CAD) model22Source code and algorithm have been made publicly available at: https://github.com/DIAGNijmegen/prostateMR_3D-CAD-csPCa https://grand-challenge.org/algorithms/prostate-mri-cad-cspca for automated localization of clinically significant prostate cancer (csPCa) in bi-parametric MR imaging (bpMRI). Deep attention mechanisms drive its detection network, targeting salient structures and highly discriminative feature dimensions across multiple resolutions. Its goal is to accurately identify csPCa lesions from indolent cancer and the wide range of benign pathology that can afflict the prostate gland. Simultaneously, a decoupled residual classifier is used to achieve consistent false positive reduction, without sacrificing high sensitivity or computational efficiency. In order to guide model generalization with domain-specific clinical knowledge, a probabilistic anatomical prior is used to encode the spatial prevalence and zonal distinction of csPCa. Using a large dataset of 1950 prostate bpMRI paired with radiologically-estimated annotations, we hypothesize that such CNN-based models can be trained to detect biopsy-confirmed malignancies in an independent cohort.For 486 institutional testing scans, the 3D CAD system achieves 83.69±5.22% and 93.19±2.96% detection sensitivity at 0.50 and 1.46 false positive(s) per patient, respectively, with 0.882±0.030 AUROC in patient-based diagnosis –significantly outperforming four state-of-the-art baseline architectures (U-SEResNet, UNet++, nnU-Net, Attention U-Net) from recent literature. For 296 external biopsy-confirmed testing scans, the ensembled CAD system shares moderate agreement with a consensus of expert radiologists (76.69%; kappa = 0.51±0.04) and independent pathologists (81.08%; kappa = 0.56±0.06); demonstrating strong generalization to histologically-confirmed csPCa diagnosis.

WDO optimized detection for mammographic masses and its diagnosis: A unified CAD system

For decades, breast cancer is the leading cause of cancer-related deaths among women. Early detection and diagnosis of breast cancer can effectively reduce the mortality rate. Masses are one of the manifestations of imperceptible breast cancer visible in mammograms whose detection and diagnosis is a challenging task due to its subtle nature. In this paper, a unified computer-aided detection/diagnosis scheme is proposed for the automatic detection and diagnosis of mammographic masses. Being simple and effective, multilevel image thresholding based on Otsu’s method is considered for localization of suspicious mass lesions, but its performance is compromised while defining the precise threshold value especially when the intensity profile in and around the anomalies does not vary much. To mitigate the issue, a nature-inspired optimization algorithm, wind driven optimization, is proposed which in combination with Otsu’s multilevel thresholding identifies the potential candidates of mass lesions. To reduce the false positives and diagnose the malignant masses, a texture-based multi-gradient local quinary pattern (M-GQP) feature is introduced which renders higher consistency in computing gradient information from the uniform and near-uniform areas. Due to eight distinct Sobel masks, it offers a better resolution in edge regions in addition to the micro information at different orientations. Four popular classifiers (support vector machine, Fisher’s linear discriminant, K-nearest neighbor, and an ensemble classifier) are also investigated to discriminate malignant tumors from benign lesions and normal tissues. The proposed integrated approach for the detection and characterization of mammographic masses is evaluated on two benchmark databases — mini-MIAS and DDSM comprising of 68 and 500 mammograms, respectively and achieved a sensitivity of 96.9% and 96.2% with 0.09 and 0.17 false positives per image after false positive reduction for the respective datasets. The diagnosis of malignant masses prior to and following false-positive reduction observes an Az value of 0.98 and 0.94 with an accuracy of 99.04% and 92.65%, respectively for the mini-MIAS dataset while the same for the DDSM are 0.99 and 0.92 with an accuracy of 98.33% and 78.50%, respectively. In a three-class classification (normal, benign and malignant), an accuracy of 99.04% and 97.76% with an F1 score of 0.98 and 0.97 for malignancy is obtained for the mini-MIAS and DDSM images, respectively which are promising in nature when compared to other competing schemes in the state-of-the-art.

Cleaning the Medicago Microarray Database to Improve Gene Function Analysis.

Transcriptomics studies have been facilitated by the development of microarray and RNA-Seq technologies, with thousands of expression datasets available for many species. However, the quality of data can be highly variable, making the combined analysis of different datasets difficult and unreliable. Most of the microarray data for Medicago truncatula, the barrel medic, have been stored and made publicly accessible on the web database Medicago truncatula Gene Expression atlas (MtGEA). The aim of this work is to ameliorate the quality of the MtGEA database through a general method based on logical and statistical relationships among parameters and conditions. The initial 716 columns available in the dataset were reduced to 607 by evaluating the quality of data through the sum of the expression levels over the entire transcriptome probes and Pearson correlation among hybridizations. The reduced dataset shows great improvements in the consistency of the data, with a reduction in both false positives and false negatives resulting from Pearson correlation and GO enrichment analysis among genes. The approach we used is of general validity and our intent is to extend the analysis to other plant microarray databases.