Computer-aided Detection Algorithm Research Articles

Improving lesion detection in mammograms by leveraging a Cycle-GAN-based lesion remover

BackgroundThe wide heterogeneity in the appearance of breast lesions and normal breast structures can confuse computerized detection algorithms. Our purpose was therefore to develop a Lesion Highlighter (LH) that can improve the performance of computer-aided detection algorithms for detecting breast cancer on screening mammograms.MethodsWe hypothesized that a Cycle-GAN based Lesion Remover (LR) could act as an LH, which can improve the performance of lesion detection algorithms. We used 10,310 screening mammograms from 4,832 women that included 4,942 recalled lesions (BI-RADS 0) and 5,368 normal results (BI-RADS 1). We divided the dataset into Train:Validate:Test folds with the ratios of 0.64:0.16:0.2. We segmented image patches (400 × 400 pixels) from either lesions marked by MQSA radiologists or normal tissue in mammograms. We trained a Cycle-GAN to develop two GANs, where each GAN transferred the style of one image to another. We refer to the GAN transferring the style of a lesion to normal breast tissue as the LR. We then highlighted the lesion by color-fusing the mammogram after applying the LR to its original. Using ResNet18, DenseNet201, EfficientNetV2, and Vision Transformer as backbone architectures, we trained three deep networks for each architecture, one trained on lesion highlighted mammograms (Highlighted), another trained on the original mammograms (Baseline), and Highlighted and Baseline combined (Combined). We conducted ROC analysis for the three versions of each deep network on the test set.ResultsThe Combined version of all networks achieved AUCs ranging from 0.963 to 0.974 for identifying the image with a recalled lesion from a normal breast tissue image, which was statistically improved (p-value < 0.001) over their Baseline versions with AUCs that ranged from 0.914 to 0.967.ConclusionsOur results showed that a Cycle-GAN based LR is effective for enhancing lesion conspicuity and this can improve the performance of a detection algorithm.

Tweaking algorithms. Technopolitical issues associated with artificial intelligence based tuberculosis detection in global health.

Computer-aided detection algorithms based on artificial intelligence are increasingly being tested and used as a means for detecting tuberculosis in countries where the epidemic is still present. Computer-aided detection tools are often presented as a global solution that can be deployed in all the geographical areas concerned by tuberculosis, but at the same time, they need to be adjusted and calibrated according to local populations' characteristics. The aim of this article is to analyze the tensions between the standardization of computer-aided detection algorithms and their local adaptation and the political issues associated with these tensions. We undertook a qualitative analysis of practices associated with tuberculosis detection algorithms in different contexts, contrasting the perspectives of various stakeholders. Algorithms embed the promise of standardization through automation and the bypassing of variable human expertise such as that of radiologists, they are nonetheless objects of local practices that we have characterized as "tweaking." This work of tweaking reveals how the technology is situated but also the many concerns of the users and workers (insertion in care, control over infrastructure, and political ownership). This should be better considered to truly make computer-aided detection innovative tools for tuberculosis management in global health.

Localization model and rank-based features selection approach for the classification of GGO and consolidation stages of COVID-19

Diseases like Middle East respiratory syndrome (MERS) and severe acute respiratory syndrome (SARS) have been caused by a large family virus known as Coronavirus. High-resolution computed tomography (HRCT) chest is used as a useful tool for the triage and stratification of COVID-19-suspected patients having respiratory symptoms until they are confirmed by polymerase chain reaction (PCR) test. To further expedite the reporting of HRCT chest in a short time, computer-aided detection algorithms can be more useful. In this work, a method is proposed that consists of three phases including COVID-19 classification, localization, and segmentation. The COVID-19-ResNet-50 model is designed for COVID-19 classification using HRCT slices. The features are obtained from a pre-trained ResNet-50 model, and the process of identifying the most pertinent features is carried out using the Wilcoxon Separability Criteria (WSC). The linear discriminant (LD) classifier is used to classify COVID-19 providing an accuracy of 0.99. To localize the COVID-19-infected region, the classified HRCT slices are fed as input to an open neural network exchange (ONNX) that is a backbone of the Tiny-You Look at Once (YOLO)-v2 model. The proposed COVID-19-YOLOv2 model localized the COVID-19 lesions with 0.948 intersections over union (IoU). A three-dimensional semantic neural network comprising 10 selected layers is proposed whose training based on the best-selected hyperparameters is performed, it accurately segments the localized infected regions having an 0.98 IoU. The proposed method is accessed on national and international datasets that provide improved performance as compared to the most recent works published in this domain.

Evaluation of A Computer-Aided Detection Software for Prostate Cancer Prediction: Excellent Diagnostic Accuracy Independent of Preanalytical Factors

Prostate cancer (PCa) is the most common noncutaneous cancer in men in the Western world. In addition to accurate diagnosis, Gleason grading and tumor volume estimates are critical for patient management. Computer-aided detection (CADe) software can be used to facilitate the diagnosis and improve the diagnostic accuracy and reporting consistency. However, preanalytical factors such as fixation and staining of prostate biopsy specimens and whole slide images (WSI) on scanners can vary significantly between pathology laboratories and may, therefore, impact the quality of WSI and utility of CADe algorithms. We evaluated the performance of a CADe software in predicting PCa on WSIs of prostate biopsy specimens and focused on whether there were any significant differences in image quality between WSIs obtained on different scanners and specimens from different histopathology laboratories. Thirty prostate biopsy specimens from 2 histopathology laboratories in the United States were included in this study. The hematoxylin and eosin slides of the biopsy specimens were scanned on 3 scanners, generating 90 WSIs. These WSIs were then analyzed using a CADe software (INIFY Prostate, Algorithm), which identified and annotated all areas suspicious for PCa and calculated the tumor volume (percentage area of the biopsy core involved). Study pathologists then reviewed the Algorithm’s annotations and tumor volume calculation to confirm the diagnosis and identify benign glands that were misclassified as cancer (false positive) and cancer glands that were misclassified as benign (false negative). The CADe software worked equally well on WSIs from all 3 scanners and from both laboratories, with similar sensitivity and specificity. The overall sensitivity was 99.4%, and specificity was 97%. The percentage of suspicious cancer areas calculated by the Algorithm was similar for all 3 scanners. For WSIs with small foci of cancer (<1 mm), the Algorithm identified all cancer glands (sensitivity, 100%). Preanalytical factors had no significant impact on whole slide imaging and subsequent application of a CADe software. Prediction accuracy could potentially be further improved by processing biopsy specimens in a centralized histology laboratory and training the Algorithm on WSIs from the same laboratory in order to minimize variations in preanalytical factors and optimize the diagnostic performance of the Algorithm.

Observer Performance Evaluation of a Deep Learning Model for Multilabel Classification of Active Tuberculosis Lung Zone-Wise Manifestations.

Background Chest X-rays (CXRs) are widely used for cost-effective screening of active pulmonary tuberculosis despite their limitations in sensitivity and specificity when interpreted by clinicians or radiologists. To address this issue, computer-aided detection (CAD) algorithms, particularly deep learning architectures based on convolution, have been developed to automate the analysis of radiography imaging. Deep learning algorithms have shown promise in accurately classifying lung abnormalities using chest X-ray images. In this study, we utilized the EfficientNet B4 model, which was pre-trained on ImageNet with 380x380 input dimensions, using its weights for transfer learning, and was modified with a series of components including global average pooling, batch normalization, dropout, and a classifier with 12 image-wise and 44 segment-wise lung zone evaluation classes using sigmoid activation. Objectives Assess the clinical usefulness of our previously created EfficientNet B4 model in identifying lung zone-specific abnormalities related to active tuberculosis through an observer performance test involving a skilled clinician operating in tuberculosis-specific environments. Methods The ground truth was established by a radiologist who examined all sample CXRs to identify lung zone-wise abnormalities. An expert clinician working in tuberculosis-specific settings independently reviewed the same CXR with blinded access to the ground truth. Simultaneously, the CXRs were classified using the EfficientNet B4 model. The clinician's assessments were then compared with the model's predictions, and the agreement between the two was measured using the kappa coefficient, evaluating the model's performance in classifying active tuberculosis manifestations across lung zones. Results The results show a strong agreement (Kappa ≥0.81) seen for lung zone-wise abnormalities of pneumothorax, mediastinal shift, emphysema, fibrosis, calcifications, pleural effusion, and cavity. Substantial agreement (Kappa = 0.61-0.80) for cavity, mediastinal shift, volume loss, and collapsed lungs. The Kappa score for lung zone-wise abnormalities is moderate (0.41-0.60) for 39% of cases. In image-wise agreement, the EfficientNet B4 model's performance ranges from moderate to almost perfect across categories, while in lung zone-wise agreement, it varies from fair to almost perfect. The results show strongagreement between the EfficientNet B4 model and the human reader in detecting lung zone-wise and image-wise manifestations. Conclusion The clinical utility of the EfficientNet B4 models to detect the abnormalities can aid clinicians in primary care settings for screening and triaging tuberculosis where resources are constrained or overburdened.

Assessment of non-tuberculosis abnormalities on digital chest x-rays with high CAD4TB scores from a tuberculosis prevalence survey in Zambia and South Africa

BackgroundChest X-rays (CXRs) have traditionally been used to aid the diagnosis of TB-suggestive abnormalities. Using Computer-Aided Detection (CAD) algorithms, TB risk is quantified to assist with diagnostics. However, CXRs capture all other structural abnormalities. Identification of non-TB abnormalities in individuals with CXRs that have high CAD scores but don’t have bacteriologically confirmed TB is unknown. This presents a missed opportunity of extending novel CAD systems’ potential to simultaneously provide information on other non-TB abnormalities alongside TB. This study aimed to characterize and estimate the prevalence of non-TB abnormalities on digital CXRs with high CAD4TB scores from a TB prevalence survey in Zambia and South Africa.MethodologyThis was a cross-sectional analysis of clinical data of participants from the TREATS TB prevalence survey conducted in 21 communities in Zambia and South Africa. The study included individuals aged ≥ 15 years who had high CAD4TB scores (score ≥ 70), but had no bacteriologically confirmed TB in any of the samples submitted, were not on TB treatment, and had no history of TB. Two consultant radiologists reviewed the images for non-TB abnormalities.ResultsOf the 525 CXRs reviewed, 46.7% (245/525) images were reported to have non-TB abnormalities. About 11.43% (28/245) images had multiple non-TB abnormalities, while 88.67% (217/245) had a single non-TB abnormality. The readers had a fair inter-rater agreement (r = 0.40). Based on anatomical location, non-TB abnormalities in the lung parenchyma (19%) were the most prevalent, followed by Pleura (15.4%), then heart & great vessels (6.1%) abnormalities. Pleural effusion/thickening/calcification (8.8%) and cardiomegaly (5%) were the most prevalent non-TB abnormalities. Prevalence of (2.7%) for pneumonia not typical of pulmonary TB and (2.1%) mass/nodules (benign/ malignant) were also reported.ConclusionA wide range of non-TB abnormalities can be identified on digital CXRs among individuals with high CAD4TB scores but don’t have bacteriologically confirmed TB. Adaptation of AI systems like CAD4TB as a tool to simultaneously identify other causes of abnormal CXRs alongside TB can be interesting and useful in non-faculty-based screening programs to better link cases to appropriate care.

Automated Detection of Periodontal Bone Loss Using Deep Learning and Panoramic Radiographs: A Convolutional Neural Network Approach

(1) Background: The accurate diagnosis of periodontal disease typically involves complex clinical and radiologic examination. However, recent studies have demonstrated the potential of deep learning in improving diagnostic accuracy and reliability through the development of computer-aided detection and diagnosis algorithms for dental problems using various radiographic sources. This study focuses on the use of panoramic radiographs, which are preferred due to their ability to assess the entire dentition with a single radiation dose. The objective is to evaluate whether panoramic radiographs are a reliable source for the detection of periodontal bone loss using deep learning, and to assess its potential for practical use on a large dataset. (2) Methods: A total of 4083 anonymized digital panoramic radiographs were collected using a Proline XC machine (Planmeca Co., Helsinki, Finland) in accordance with the research ethics protocol. These images were used to train the Faster R-CNN object detection method for detecting periodontally compromised teeth on panoramic radiographs. (3) Results: This study demonstrated a high level of consistency and reproducibility among examiners, with overall inter- and intra-examiner correlation coefficient (ICC) values of 0.94. The Area Under the Curve (AUC) for detecting periodontally compromised and healthy teeth was 0.88 each, and the overall AUC for the entire jaw, including edentulous regions, was 0.91. (4) Conclusions: The regional grouping of teeth exhibited reliable detection performance for periodontal bone loss using a large dataset, indicating the possibility of automating the diagnosis of periodontitis using panoramic radiographs.

Mammography and ultrasound based dual modality classification of breast cancer using a hybrid deep learning approach

Traditional methods of diagnosing breast cancer (BC) suffer from human errors, are less accurate, and consume time. A computer-aided detection (CAD) system can overcome the above-stated limitations and help radiologists with accurate decision-making. However, the existing studies using single imaging modalities have shown limited clinical use due to its low diagnostic accuracy and reliability when compared to multimodal system. Thus, we aim to develop a hybrid deep learning bimodal CAD algorithm for the classification of breast lesions using mammogram and ultrasound imaging modalities combined. A combined convolutional neural network (CNN) and long-short term memory (LSTM) model is implemented using images from both mammogram and ultrasound modalities to improve the early diagnosis of BC. A new real-time dataset consisting of 43 mammogram images and 43 ultrasound images collected from 31 patients is used in this work. Further, each group consists of 25 benign and 18 malignant images. The number of images is increased to 1032 (516 for each modality) using different data augmentation techniques. The proposed bimodal CAD algorithm achieves a classification accuracy of 99.35% and the area under the receiver operating characteristic curve (AUC) of 0.99 over the traditional unimodal CAD systems, which attain the classification accuracy of 97.16% and 98.84% using mammogram and ultrasound, respectively. The proposed bimodal CAD algorithm using combined mammogram and ultrasound outperforms the traditional unimodal CAD systems. The bimodal CAD algorithm can avoid unnecessary biopsies and encourage its clinical application.

Integrated IoT-based Healthcare System for the Early Detection of Breast Cancer Using Intelligent Diagnostic System

Breast cancer represents one of the leading cancer-related diseases worldwide, affecting mostly women after puberty. Even though the illness is fatal and kills thousands of people each year, it is mostly curative if found quickly. As a result, prompt and precise detection methods are critical to patient survival. Previously, doctors used manual detection systems for this objective. However, such techniques have been slow and frequently dependent on the physician's expertise. As technology advanced, these primitive methodologies were supplemented by computer-aided detection (CAD) algorithms. Deep learning is extremely common because of the massive development in large data, the Internet of Things (IoT), linked devices, and high-performance computers using GPUs and TPUs. The Internet of Things (IoT) has advanced recently, and the healthcare industry is benefiting from this growth. Sensors that gather data for required analysis are crucial components utilized in the Internet of Things. Physicians and medical staff will be able to carry out their tasks with ease and intelligence thanks to the Internet of Things. The proposed research focus on integrating Alexnet and ResNet101 for accurate prediction of Breast malignancy from mammogram data. This methodology will target the features more precisely than any other combination of the pre-trained model. Finally, to resolve the computational burden issue, the feature reduction ReliefF methodology is used. To demonstrate the proposed method, an online publicly released set of data of 750 BU images is used. For training and testing the models, the set of data has been further split into 80 and 20% ratios. Following extensive testing and analysis, it was discovered that the DenseNet-201 and MobileNet-v2 trained SVMs to have an accuracy of 98.39 percent for the original and augmented Mammo images online datasets, respectively. This research discovered that the proposed approach is efficient and simple to implement to assist radiographers and physicians in diagnosing breast cancer in females.

Practice toward standardized performance testing of computer-aided detection algorithms for pulmonary nodule.

This study aimed at implementing practice to build a standardized protocol to test the performance of computer-aided detection (CAD) algorithms for pulmonary nodules. A test dataset was established according to a standardized procedure, including data collection, curation and annotation. Six types of pulmonary nodules were manually annotated as reference standard. Three specific rules to match algorithm output with reference standard were applied and compared. These rules included: (1) "center hit" [whether the center of algorithm highlighted region of interest (ROI) hit the ROI of reference standard]; (2) "center distance" (whether the distance between algorithm highlighted ROI center and reference standard center was below a certain threshold); (3) "area overlap" (whether the overlap between algorithm highlighted ROI and reference standard was above a certain threshold). Performance metrics were calculated and the results were compared among ten algorithms under test (AUTs). The test set currently consisted of CT sequences from 593 patients. Under "center hit" rule, the average recall rate, average precision, and average F1 score of ten algorithms under test were 54.68, 38.19, and 42.39%, respectively. Correspondingly, the results under "center distance" rule were 55.43, 38.69, and 42.96%, and the results under "area overlap" rule were 40.35, 27.75, and 31.13%. Among the six types of pulmonary nodules, the AUTs showed the highest miss rate for pure ground-glass nodules, with an average of 59.32%, followed by pleural nodules and solid nodules, with an average of 49.80 and 42.21%, respectively. The algorithm testing results changed along with specific matching methods adopted in the testing process. The AUTs showed uneven performance on different types of pulmonary nodules. This centralized testing protocol supports the comparison between algorithms with similar intended use, and helps evaluate algorithm performance.

A Fully Deep Learning Paradigm for Pneumoconiosis Staging on Chest Radiographs.

Pneumoconiosis staging has been a very challenging task, both for certified radiologists and computer-aided detection algorithms. Although deep learning has shown proven advantages in the detection of pneumoconiosis, it remains challenging in pneumoconiosis staging due to the stage ambiguity of pneumoconiosis and noisy samples caused by misdiagnosis when they are used in training deep learning models. In this article, we propose a fully deep learning pneumoconiosis staging paradigm that comprises a segmentation procedure and a staging procedure. The segmentation procedure extracts lung fields in chest radiographs through an Asymmetric Encoder-Decoder Network (AED-Net) that can mitigate the domain shift between multiple datasets. The staging procedure classifies the lung fields into four stages through our proposed deep log-normal label distribution learning and focal staging loss. The two cascaded procedures can effectively solve the problem of model overfitting caused by stage ambiguity and noisy labels of pneumoconiosis. Besides, we collect a clinical chest radiograph dataset of pneumoconiosis from the certified radiologist's diagnostic reports. The experimental results on this novel pneumoconiosis dataset confirm that the proposed deep pneumoconiosis staging paradigm achieves an Accuracy of 90.4%, a Precision of 84.8%, a Sensitivity of 78.4%, a Specificity of 95.6%, an F1-score of 80.9% and an Area Under the Curve (AUC) of 96%. In particular, we achieve 68.4% Precision, 76.5% Sensitivity, 95% Specificity, 72.2% F1-score and 89% AUC on the early pneumoconiosis 'stage-1'.

Data enhancement based on M2-Unet for liver segmentation in Computed Tomography

Liver cancer is one of the diseases with the highest mortality in the world. Accurate segmentation of liver regions from liver CT data is a necessary preprocessing stage of computer-aided detection and diagnosis algorithms. In view of the small number of existing medical image data sets and difficulty in acquisition, this paper expanded the effective data set based on an improved conditional generation adversarial network pix2pix. We added random noise to the input of the generator, and the last layer of the output of the discriminator. Moreover, the original fully connected layer was replaced with a probability matrix with a size of 16 × 16. This is to make the discriminator more delicate. We merged the block module in MobileNet-v2 and U-Net (M2-Unet) to achieve liver segmentation. In addition to the training and validation using the public competition data set LiTs, we also constructed a new liver data set. The Dice similarity coefficient of the algorithm in this paper is 88.7%. The proposed method has improved to the ordinary U-Net algorithm. In the test phase, each image only needs 0.072 s to complete the segmentation on average, which exceeds the segmentation speed of specialists. Experimental results show that algorithm in this paper can accurately segment the liver and meet the requirements of realtime segmentation.

3cDe-Net: a cervical cancer cell detection network based on an improved backbone network and multiscale feature fusion

BackgroundCervical cancer cell detection is an essential means of cervical cancer screening. However, for thin-prep cytology test (TCT)-based images, the detection accuracies of traditional computer-aided detection algorithms are typically low due to the overlapping of cells with blurred cytoplasmic boundaries. Some typical deep learning-based detection methods, e.g., ResNets and Inception-V3, are not always efficient for cervical images due to the differences between cervical cancer cell images and natural images. As a result, these traditional networks are difficult to directly apply to the clinical practice of cervical cancer screening.MethodWe propose a cervical cancer cell detection network (3cDe-Net) based on an improved backbone network and multiscale feature fusion; the proposed network consists of the backbone network and a detection head. In the backbone network, a dilated convolution and a group convolution are introduced to improve the resolution and expression ability of the model. In the detection head, multiscale features are obtained based on a feature pyramid fusion network to ensure the accurate capture of small cells; then, based on the Faster region-based convolutional neural network (R-CNN), adaptive cervical cancer cell anchors are generated via unsupervised clustering. Furthermore, a new balanced L1-based loss function is defined, which reduces the unbalanced sample contribution loss.ResultBaselines including ResNet-50, ResNet-101, Inception-v3, ResNet-152 and the feature concatenation network are used on two different datasets (the Data-T and Herlev datasets), and the final quantitative results show the effectiveness of the proposed dilated convolution ResNet (DC-ResNet) backbone network. Furthermore, experiments conducted on both datasets show that the proposed 3cDe-Net, based on the optimal anchors, the defined new loss function, and DC-ResNet, outperforms existing methods and achieves a mean average precision (mAP) of 50.4%. By performing a horizontal comparison of the cells on an image, the category and location information of cancer cells can be obtained concurrently.ConclusionThe proposed 3cDe-Net can detect cancer cells and their locations on multicell pictures. The model directly processes and analyses samples at the picture level rather than at the cellular level, which is more efficient. In clinical settings, the mechanical workloads of doctors can be reduced, and their focus can be placed on higher-level review work.

Strengths and Weaknesses of an Artificial Intelligence Polyp Detection Program as Assessed by a High-Detecting Endoscopist

Strengths and Weaknesses of an Artificial Intelligence Polyp Detection Program as Assessed by a High-Detecting Endoscopist

Bayesian spatial models for voxel-wise prostate cancer classification using multi-parametric magnetic resonance imaging data.

Multi-parametric magnetic resonance imaging (mpMRI) has been playing an increasingly important role in the detection of prostate cancer (PCa). Various computer-aided detection algorithms were proposed for automated PCa detection by combining information in multiple mpMRI parameters. However, there are specific features of mpMRI, including between-voxel correlation within each prostate and heterogeneity across patients, that have not been fully explored but could potentially improve PCa detection if leveraged appropriately. This article proposes novel Bayesian approaches for voxel-wise PCa classification that accounts for spatial correlation and between-patient heterogeneity in the mpMRI data. Modeling the spatial correlation is challenging due to the extreme high dimensionality of the data, and we propose three scalable approaches based on Nearest Neighbor Gaussian Process (NNGP), reduced-rank approximation, and a conditional autoregressive (CAR) model that approximates a Gaussian Process with the Matérn covariance, respectively. Our simulation study shows that properly modeling the spatial correlation and between-patient heterogeneity can substantially improve PCa classification. Application to in vivo data illustrates that classification is improved by all three spatial modeling approaches considered, while modeling the between-patient heterogeneity does not further improve our classifiers. Among the proposed models, the NNGP-based model is recommended given its high classification accuracy and computational efficiency.

Texture-based classification of lung disease patterns in chronic hypersensitivity pneumonitis and comparison to clinical outcomes.

Computer-aided detection algorithms applied to CT lung imaging have the potential to objectively quantify pulmonary pathology. We aim to develop an automatic classification method based on textural features able to classify healthy and pathological patterns on CT lung images and to quantify the extent of each disease pattern in a group of patients with chronic hypersensitivity pneumonitis (cHP), in comparison to pulmonary function tests (PFTs).27 cHP patients were scanned via high resolution CT (HRCT) at full-inspiration. Regions of interest (ROIs) were extracted and labeled as normal (NOR), ground glass opacity (GGO), reticulation (RET), consolidation (C), honeycombing (HB) and air trapping (AT). For each ROI, statistical, morphological and fractal parameters were computed. For automatic classification, we compared two classification methods (Bayesian and Support Vector Machine) and three ROI sizes. The classifier was therefore applied to the overall CT images and the extent of each class was calculated and compared to PFTs. Better classification accuracy was found for the Bayesian classifier and the 16x16 ROI size: 92.1±2.7%. The extent of GGO, HB and NOR significantly correlated with forced vital capacity (FVC) and the extent of NOR with carbon monoxide diffusing capacity (DLCO).Clinical Relevance- Texture analysis can differentiate and objectively quantify pathological classes in the lung parenchyma and may represent a quantitative diagnostic tool in cHP.

An AI-Based Algorithm for the Automatic Classification of Thoracic Radiographs in Cats.

An artificial intelligence (AI)-based computer-aided detection (CAD) algorithm to detect some of the most common radiographic findings in the feline thorax was developed and tested. The database used for training comprised radiographs acquired at two different institutions. Only correctly exposed and positioned radiographs were included in the database used for training. The presence of several radiographic findings was recorded. Consequenly, the radiographic findings included for training were: no findings, bronchial pattern, pleural effusion, mass, alveolar pattern, pneumothorax, cardiomegaly. Multi-label convolutional neural networks (CNNs) were used to develop the CAD algorithm, and the performance of two different CNN architectures, ResNet 50 and Inception V3, was compared. Both architectures had an area under the receiver operating characteristic curve (AUC) above 0.9 for alveolar pattern, bronchial pattern and pleural effusion, an AUC above 0.8 for no findings and pneumothorax, and an AUC above 0.7 for cardiomegaly. The AUC for mass was low (above 0.5) for both architectures. No significant differences were evident in the diagnostic accuracy of either architecture.

S212 Comparing Polyp Detection Rate Between Water Exchange and Air Insufflation by a Computer-aided Detection Algorithm: An Analysis of Withdrawal Phase Videos from a Randomized Controlled Trial

Introduction: Water exchange (WE) improves polyp (PDR) and adenoma detection rate (ADR) compared with air insufflation [GIE 2021;93:1411], but still misses polyps due to human limitations [JCG 2021, in press]. A computer-aided detection (CADe) algorithm can overcome human omissions, but is limited by false positives (FPs) [Lancet Gastroenterol & Hepatol 2019;4:71], which might be reduced by salvage-cleaning effect provided by insertion WE [Tzu-Chi Med J 2021;33:108]. We hypothesized that the strengths of CADe and WE can complement the weaknesses of each other. Their combination can enhance PDR and ADR and reduce false positive rate (FPR). Methods: We used a locally developed convolutional neural network with YOLOv4 to build a CADe model. Images of polyps (n=15961) were divided in 80% and 20% for training and testing. Prior to real-time application, we validated the model in detecting polyps, using edited and coded withdrawal phase videos of the right colon from a RCT comparing right colon adenoma detection between WE and air insufflation. The polyps detected by the CADe were confirmed by the consensus of 2 blinded reviewers (experienced endoscopists). We also developed a computer-aided diagnosis (CADx) model based on YOLOv4 to characterize the detected polyps as adenomas or not. The primary and secondary outcomes were PDR and the FPR, respectively. Results: A total of 245 coded videos of colonoscopies inserted with WE (n=123) or air insufflation (n=122) were analyzed. CADe achieved significantly higher PDR in the WE group (61.8%) than in the air insufflation group (40.2%), P=0.0009. CADe also had significantly higher additional PDR in the WE group (30.1% vs 12.3%, P=0.001). The CADx assessed ADR was significantly higher in the WE group than that in the air insufflation group (46.3% vs 30.3%, P=0.018), after the detected polyps were characterized by the CADx. (Table a) The mean numbers of FPs related to feces [1.78 (1.67) vs 2.09 (2.09), P=0.007] and bubbles [0.53 (0.89) vs 1.25 (2.45), P=0.001] were significantly lower in the WE group than in the air insufflation group. (Table b). Conclusion: The data confirmed the hypothesis that the strengths of CADe and WE complemented the weaknesses of each other, enhanced PDR and ADR and reduced FPR. The locally developed CADe and CADx are validated for polyp detection and adenoma assignment for real-time application. Furthermore, WE provides an optimal platform for the computer-assisted algorithms.Table 1.: ADR, adenoma detection rate; All the percentages calculated according to respective total number of patients for each group; PDR, polyp detection rate; Total numbers of polyps were based on those found by the performing endoscopist plus those detected by the CADe; Proportion of additional polyps found by CADe equals numbers of additional polyp (adenomas) found by CADe divided by total numbers of polyps (adenomas). Others: ileocecal valve, cecum, polypectomy site, suction marks and drug pills. † Mann-Whitney test; ††Fisher exact test.

Present Limitations of Artificial Intelligence in the Emergency Setting - Performance Study of a Commercial, Computer-Aided Detection Algorithm for Pulmonary Embolism.

Since artificial intelligence is transitioning from an experimental stage to clinical implementation, the aim of our study was to evaluate the performance of a commercial, computer-aided detection algorithm of computed tomography pulmonary angiograms regarding the presence of pulmonary embolism in the emergency room. This retrospective study includes all pulmonary computed tomography angiogram studies performed in a large emergency department over a period of 36months that were analyzed by two radiologists experienced in emergency radiology to set a reference standard. Original reports and computer-aided detection results were compared regarding the detection of lobar, segmental, and subsegmental pulmonary embolism. All computer-aided detection findings were analyzed concerning the underlying pathology. False-positive findings were correlated to the contrast-to-noise ratio. Expert reading revealed pulmonary embolism in 182 of 1229 patients (49 % men, 10-97 years) with a total of 504emboli. The computer-aided detection algorithm reported 3331 findings, including 258 (8 %) true-positive findings and 3073 (92 %) false-positive findings. Computer-aided detection analysis showed a sensitivity of 47 % (95 %CI: 33-61 %) on the lobar level and 50 % (95 %CI 43-56 %) on the subsegmental level. On average, there were 2.25 false-positive findings per study (median 2, range 0-25). There was no significant correlation between the number of false-positive findings and the contrast-to-noise ratio (Spearman's Rank Correlation Coefficient = 0.09). Soft tissue (61.0 %) and pulmonary veins (24.1 %) were the most common underlying reasons for false-positive findings. Applied to a population at a large emergency room, the tested commercial computer-aided detection algorithm faced relevant performance challenges that need to be addressed in future development projects. · Computed tomography pulmonary angiograms are frequently acquired in emergency radiology.. · Computer-aided detection algorithms (CADs) can support image analysis.. · CADs face challenges regarding false-positive and false-negative findings.. · Radiologists using CADs need to be aware of these limitations.. · Further software improvements are necessary ahead of implementation in the daily routine.. · Müller-Peltzer K, Kretzschmar L, Negrão de Figueiredo G et al. Present Limitations of Artificial Intelligence in the Emergency Setting - Performance Study of a Commercial, Computer-Aided Detection Algorithm for Pulmonary Embolism. Fortschr Röntgenstr 2021; 193: 1436 - 1444.

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.