Computer-aided Detection Tool Research Articles

Deep learning for colorectal cancer detection in contrast-enhanced CT without bowel preparation: a retrospective, multicentre study

Contrast-enhanced CT scans provide a means to detect unsuspected colorectal cancer. However, colorectal cancers in contrast-enhanced CT without bowel preparation may elude detection by radiologists. We aimed to develop a deep learning (DL) model for accurate detection of colorectal cancer, and evaluate whether it could improve the detection performance of radiologists. We developed a DL model using a manually annotated dataset (1196 cancer vs 1034 normal). The DL model was tested using an internal test set (98 vs 115), two external test sets (202 vs 265 in 1, and 252 vs 481 in 2), and a real-world test set (53 vs 1524). We compared the detection performance of the DL model with radiologists, and evaluated its capacity to enhance radiologists' detection performance. In the four test sets, the DL model had the area under the receiver operating characteristic curves (AUCs) ranging between 0.957 and 0.994. In both the internal test set and external test set 1, the DL model yielded higher accuracy than that of radiologists (97.2% vs 86.0%, p<0.0001; 94.9% vs 85.3%, p<0.0001), and significantly improved the accuracy of radiologists (93.4% vs 86.0%, p<0.0001; 93.6% vs 85.3%, p<0.0001). In the real-world test set, the DL model delivered sensitivity comparable to that of radiologists who had been informed about clinical indications for most cancer cases (94.3% vs 96.2%, p>0.99), and it detected 2 cases that had been missed by radiologists. The developed DL model can accurately detect colorectal cancer and improve radiologists' detection performance, showing its potential as an effective computer-aided detection tool. This study was supported by National Science Fund for Distinguished Young Scholars of China (No. 81925023); Regional Innovation and Development Joint Fund ofNational Natural Science Foundation of China(No.U22A20345);National Natural Science Foundation of China (No. 82072090 and No. 82371954); Guangdong Provincial Key Laboratory of Artificial Intelligence in Medical Image Analysis and Application (No. 2022B1212010011); High-level Hospital Construction Project (No. DFJHBF202105).

Performance of Automated ASPECTS Software and Value as a Computer-Aided Detection Tool.

ASPECTS quantifies early ischemic changes in anterior circulation stroke on NCCT but has interrater variability. We examined the agreement of conventional and automated ASPECTS and studied the value of computer-aided detection. We retrospectively collected imaging data from consecutive patients with acute ischemic stroke with large-vessel occlusion undergoing thrombectomy. Five raters scored conventional ASPECTS on baseline NCCTs, which were also processed by RAPID software. Conventional and automated ASPECTS were compared with a consensus criterion standard. We determined the agreement over the full ASPECTS range as well as dichotomized, reflecting thrombectomy eligibility according to the guidelines (ASPECTS 0-5 versus 6-10). Raters subsequently scored ASPECTS on the same NCCTs with assistance of the automated ASPECTS outputs, and agreement was obtained. For the total of 175 cases, agreement among raters individually and the criterion standard varied from fair to good (weighted κ = between 0.38 and 0.76) and was moderate (weighted κ = 0.59) for the automated ASPECTS. The agreement of all raters individually versus the criterion standard improved with software assistance, as did the interrater agreement (overall Fleiss κ = 0.15-0.23; P < .001 and .39 to .55; P = .01 for the dichotomized ASPECTS). Automated ASPECTS had agreement with the criterion standard similar to that of conventional ASPECTS. However, including automated ASPECTS during the evaluation of NCCT in acute stroke improved the agreement with the criterion standard and improved interrater agreement, which could, therefore, result in more uniform scoring in clinical practice.

Detection of pancreatic cancer with two- and three-dimensional radiomic analysis in a nationwide population-based real-world dataset

BackgroundCT is the major detection tool for pancreatic cancer (PC). However, approximately 40% of PCs < 2 cm are missed on CT, underscoring a pressing need for tools to supplement radiologist interpretation.MethodsContrast-enhanced CT studies of 546 patients with pancreatic adenocarcinoma diagnosed by histology/cytology between January 2005 and December 2019 and 733 CT studies of controls with normal pancreas obtained between the same period in a tertiary referral center were retrospectively collected for developing an automatic end-to-end computer-aided detection (CAD) tool for PC using two-dimensional (2D) and three-dimensional (3D) radiomic analysis with machine learning. The CAD tool was tested in a nationwide dataset comprising 1,477 CT studies (671 PCs, 806 controls) obtained from institutions throughout Taiwan.ResultsThe CAD tool achieved 0.918 (95% CI, 0.895–0.938) sensitivity and 0.822 (95% CI, 0.794–0.848) specificity in differentiating between studies with and without PC (area under curve 0.947, 95% CI, 0.936–0.958), with 0.707 (95% CI, 0.602–0.797) sensitivity for tumors < 2 cm. The positive and negative likelihood ratios of PC were 5.17 (95% CI, 4.45–6.01) and 0.10 (95% CI, 0.08–0.13), respectively. Where high specificity is needed, using 2D and 3D analyses in series yielded 0.952 (95% CI, 0.934–0.965) specificity with a sensitivity of 0.742 (95% CI, 0.707–0.775), whereas using 2D and 3D analyses in parallel to maximize sensitivity yielded 0.915 (95% CI, 0.891–0.935) sensitivity at a specificity of 0.791 (95% CI, 0.762–0.819).ConclusionsThe high accuracy and robustness of the CAD tool supported its potential for enhancing the detection of PC.

Development of a computer-aided tool for detection of COVID-19 pneumonia from CXR images using machine learning algorithm

The novel coronavirus (SARS-CoV-2) is spreading rapidly worldwide, and it has become a greater risk for human beings. To curb the community transmission of this virus, rapid detection and identification of the affected people via a quick diagnostic process are necessary. Media studies have shown that most COVID-19 victims endure lung disease. For rapid identification of the affected patient, chest CT scans and X-ray images have been reported to be suitable techniques. However, chest X-ray (CXR) shows more convenience than the CT imaging techniques because it has faster imaging times than CT and is also simple and cost-effective. Literature shows that transfer learning is one of the most successful techniques to analyze chest X-ray images and correctly identify various types of pneumonia. Since SVM has a remarkable aspect that tremendously provides good results using a small data set thus in this study we have used SVM machine learning algorithm to diagnose COVID-19 from chest X-ray images. The image processing tool called RGB and SqueezeNet models were used to get more images to diagnose the available data set. Our adopted model shows an accuracy of 98.8% to detect the COVID-19 affected patient from CXR images. It is expected that our proposed computer-aided detection tool (CAT) will play a key role in reducing the spread of infectious diseases in society through a faster patient screening process.

Reproducibility of lung nodule radiomic features: Multivariable and univariable investigations that account for interactions between CT acquisition and reconstruction parameters.

Recent studies have demonstrated a lack of reproducibility of radiomic features in response to variations in CT parameters. In addition, reproducibility of radiomic features has not been well established in clinical datasets. We aimed to investigate the effects of a wide range of CT acquisition and reconstruction parameters on radiomic features in a realistic setting using clinical low dose lung cancer screening cases. We performed univariable and multivariable explorations to consider the effects of individual parameters and the simultaneous interactions between three different acquisition/reconstruction parameters of radiation dose level, reconstructed slice thickness, and kernel. A cohort of 89 lung cancer screening patients were collected that each had a solid lung nodule >4mm diameter. A computational pipeline was used to perform a simulation of dose reduction of the raw projection data, collected from patient scans. This was followed by reconstruction of raw data with weighted filter back projection (wFBP) algorithm and automatic lung nodule detection and segmentation using a computer-aided detection tool. For each patient, 36 different image datasets were created corresponding to dose levels of 100%, 50%, 25%, and 10% of the original dose level, three slice thicknesses of 0.6mm, 1mm, and 2mm, as well as three reconstruction kernels of smooth, medium, and sharp. For each nodule, 226 well-known radiomic features were calculated at each image condition. The reproducibility of radiomic features was first evaluated by measuring the intercondition agreement of the feature values among the 36 image conditions. Then in a series of univariable analyses, the impact of individual CT parameters was assessed by selecting subsets of conditions with one varying and two constant CT parameters. In each subset, intraparameter agreements were assessed. Overall concordance correlation coefficient (OCCC) served as the measure of agreement. An OCCC≥0.9 implied strong agreement and reproducibility of radiomic features in intercondition or intraparameter comparisons. Furthermore, the interaction of CT parameters in impacting radiomic feature values was investigated via ANOVA. All included radiomic features lacked intercondition reproducibility (OCCC<0.9) among all the 36 conditions. Out of 226 radiomic features analyzed, only 17 and 18 features were considered reproducible (OCCC≥0.9) to dose and kernel variation, respectively, within the corresponding condition subsets. Slice thickness demonstrated the largest impact on radiomic feature values where only one to five features were reproducible at a few condition subsets. ANOVA revealed significant interactions (P<0.05) between CT parameters affecting the variability of >50% of radiomic features. We systematically explored the multidimensional space of CT parameters in affecting lung nodule radiomic features. Univariable and multivariable analyses of this study not only showed the lack of reproducibility of the majority of radiomic features but also revealed existing interactions among CT parameters, meaning that the effect of individual CT parameters on radiomic features can be conditional upon other CT acquisition and reconstruction parameters. Our findings advise on careful radiomic feature selection and attention to the inclusion criteria for CT image acquisition protocols within the datasets of radiomic studies.

Computer-aided detection tool development for teaching chest radiograph pattern recognition to undergraduate radiography students: A context needs and capability analysis.

BackgroundMedical imaging (MI) education has experienced a shift aligned with the advances in technology and the role played by radiographers in pattern recognition. This has led to increased use of technology-enhanced teaching and simulated learning approaches (e.g. computer-aided detection [CAD] tools) which also support the increasing requirement to develop pattern-recognition skills at undergraduate level. However, the development of these approaches need to be explored and planned carefully to be context-relevant.AimThe aim of this study was to explore and describe the need for and capability of a CAD tool for teaching chest radiography pattern recognition in an undergraduate radiography programme.SettingThe setting was a university that offers MI education.MethodThe study employed a qualitative descriptive design with an interpretive research paradigm. Purposive sampling was used to recruit information-rich participants for a focus group interview. Information-rich participants were considered to be those who were involved in teaching clinical skills, such as those required in pattern recognition, to radiography students. Data were transcribed verbatim and analysed in a step-by-step approach.ResultsThree main themes emerged: (1) a structured approach to enhance implicit skills is critical in the CAD tool design; (2) an authentic tool which is able to simulate real-world experiences in image analysis is essential; and (3) a tool which encourages self-directed learning using a wide variety of pathological conditions would be ideal.ConclusionThe results of this study are essential in guiding radiography educators in designing CAD tools for teaching chest radiography pattern recognition.

Intelligent Classification Technique for Breast Cancer

Breast cancer is also a leading cause of cancer death in the less developed countries of the world. This is partly because a shift in lifestyles is causing an increase in incidence. Breast cancer originates from the inner lining of milk ducts/lobes either in the form of invasive or non invasive disease in general. Mammography, particularly with Computer-Aided Detection (CAD), can now produce images detailed enough for diagnostic purposes, and digital mammography allows transmission of 3-dimensionssal images over long distances. The aim for the system is to design a Computer Aided Diagnosis systematic tool for perceiving non cancerous and perilous (cancer causing) mammogram. The aim of the research is proposed to develop an image processing algorithm for an automatic detection and classification of breast lesions accurately. CAD tool helped radiologist in expanding his assurance accuracy. Support vector machine (SVM) classifier is used to discriminate the tumors into benign or malignant. Incorporate best features of the find out that has significant responsibility in achieving the perfect turnout which are then designated and associated with ANN to train and classify.

OTHR-13. A DEEP LEARNING APPROACH TO DETECT CANCER METASTASES TO THE BRAIN IN MRI

BACKGROUND AND OBJECTIVE: Brain metastases have been found to account for one-fourth of all cancer metastases seen in clinics. Magnetic resonance imaging (MRI) is widely used for detecting brain metastases. Accurate detection of the brain metastases is critical to design radiotherapy to treat the cancer and monitor their progression or response to the therapy and prognosis. However, finding metastases on brain MRI is very challenging as many metastases are small and manifest as objects of weak contrast on the images. In this work we present a deep learning approach integrated with a classification scheme to detect cancer metastases to the brain on MRI. MATERIALS AND METHODS: We retrospectively extracted 101 metastases patients, equal to 1535 metastases on 10192 slices of images in a total of 336 scans from our PACS and manually marked the lesions on T1-weighted contrast enhanced MRI as the ground-truth. We then randomly separated the cases into training, validation, and test sets for developing and optimizing the deep learning neural network. We designed a 2-step computer-aided detection (CAD) pipeline by first applying a fast region-based convolutional neural network method (R-CNN) to sequentially process each slice of an axial brain MRI to find abnormal hyper-intensity that may correspond to a brain metastasis and, second, applying a random under sampling boost (RUSBoost) classification method to reduce the false positive metastases. RESULTS: The computational pipeline was tested on real brain images. A sensitivity of 97.28% and false positive rate of 36.25 per scan over the images were achieved by using the proposed method. CONCLUSION: Our results demonstrated the deep learning-based method can detect metastases in very challenging cases and can serve as CAD tool to help radiologists interpret brain MRIs in a time-constrained environment.

Radiological Estimation of Intracranial Blood Volume and Occurrence of Hydrocephalus Determines Stress-Induced Hyperglycemia After Aneurysmal Subarachnoid Hemorrhage.

Acute phase after aneurysmal subarachnoid hemorrhage (aSAH) is associated with several metabolic derangements including stress-induced hyperglycemia (SIH). The present study is designed to identify objective radiological determinants for SIH to better understand its contributory role in clinical outcomes after aSAH. A computer-aided detection tool was used to segment admission computed tomography (CT) images of aSAH patients to estimate intracranial blood and cerebrospinal fluid volumes. Modified Graeb score (mGS) was used as a semi-quantitative measure to estimate degree of hydrocephalus. The relationship between glycemic gap (GG) determined SIH, mGS, and estimated intracranial blood and cerebrospinal fluid volumes were evaluated using linear regression. Ninety-four [94/187 (50.3%)] among the study cohort had SIH (defined as GG > 26.7mg/dl). Patients with SIH had 14.3ml/1000ml more intracranial blood volume as compared to those without SIH [39.6ml (95% confidence interval, CI, 33.6 to 45.5) vs. 25.3ml (95% CI 20.6 to 29.9), p = 0.0002]. Linear regression analysis of mGS with GG showed each unit increase in mGS resulted in 1.2mg/dl increase in GG [p = 0.002]. Patients with SIH had higher mGS [median 4.0, interquartile range, IQR 2.0-7.0] as compared to those without SIH [median 2.0, IQR 0.0-6.0], p = 0.002. Patients with third ventricular blood on admission CT scan were more likely to develop SIH [67/118 (56.8%) vs. 27/69 (39.1%), p = 0.023]. Hence, the present study, using unbiased SIH definition and objective CT scan parameters, reports "dose-dependent" radiological features resulting in SIH. Such findings allude to a brain injury-stress response-neuroendocrine axis in etiopathogenesis of SIH.

Automated and real-time segmentation of suspicious breast masses using convolutional neural network.

In this work, a computer-aided tool for detection was developed to segment breast masses from clinical ultrasound (US) scans. The underlying Multi U-net algorithm is based on convolutional neural networks. Under the Mayo Clinic Institutional Review Board protocol, a prospective study of the automatic segmentation of suspicious breast masses was performed. The cohort consisted of 258 female patients who were clinically identified with suspicious breast masses and underwent clinical US scan and breast biopsy. The computer-aided detection tool effectively segmented the breast masses, achieving a mean Dice coefficient of 0.82, a true positive fraction (TPF) of 0.84, and a false positive fraction (FPF) of 0.01. By avoiding positioning of an initial seed, the algorithm is able to segment images in real time (13–55 ms per image), and can have potential clinical applications. The algorithm is at par with a conventional seeded algorithm, which had a mean Dice coefficient of 0.84 and performs significantly better (P< 0.0001) than the original U-net algorithm.

The use of hippocampal volumetric measurements to improve diagnostic accuracy in pediatric patients with mesial temporal sclerosis.

OBJECTIVE Many patients with medically intractable epilepsy have mesial temporal sclerosis (MTS), which significantly affects their quality of life. The surgical excision of MTS lesions can result in marked improvement or even complete resolution of the epileptic episodes. Reliable radiological diagnosis of MTS is a clinical challenge. The purpose of this study was to evaluate the utility of volumetric mapping of the hippocampi for the identification of MTS in a case-controlled series of pediatric patients who underwent resection for medically refractory epilepsy, using pathology as a gold standard. METHODS A cohort of 57 pediatric patients who underwent resection for medically intractable epilepsy between 2005 and 2015 was evaluated. On pathological investigation, this group included 24 patients with MTS and 33 patients with non-MTS findings. Retrospective quantitative volumetric measurements of the hippocampi were acquired for 37 of these 57 patients. Two neuroradiologists with more than 10 years of experience who were blinded to the patients' MTS status performed the retrospective review of MR images. To produce the volumetric data, MR scans were parcellated and segmented using the FreeSurfer software suite. Hippocampal regions of interest were compared against an age-weighted local regression curve generated with data from the pediatric normal cohort. Standard deviations and percentiles of specific subjects were calculated. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined for the original clinical read and the expert readers. Receiver operating characteristic curves were generated for the methods of classification to compare results from the readers with the authors' results, and an optimal threshold was determined. From that threshold the sensitivity, specificity, PPV, and NPV were calculated for the volumetric analysis. RESULTS With the use of quantitative volumetry, a sensitivity of 72%, a specificity of 95%, a PPV of 93%, an NPV of 78%, and an area under the curve of 0.84 were obtained using a percentage difference of normalized hippocampal volume. The resulting specificity (95%) and PPV (93%) are superior to the original clinical read and to Reader A and Reader B's findings (range for specificity 74%-86% and for PPV 64%-71%). The sensitivity (72%) and NPV (78%) are comparable to Reader A's findings (73% and 81%, respectively) and are better than those of the original clinical read and of Reader B (sensitivity 45% and 63% and NPV 71% and 70%, respectively). CONCLUSIONS Volumetric measurement of the hippocampi outperforms expert readers in specificity and PPV, and it demonstrates comparable to superior sensitivity and NPV. Volumetric measurements can complement anatomical imaging for the identification of MTS, much like a computer-aided detection tool would. The implementation of this approach in the daily clinical workflow could significantly improve diagnostic accuracy.

Computer‐aided detection and quantification of cavitary tuberculosis from CT scans

To present a computer-aided detection tool for identifying, quantifying, and evaluating tuberculosis (TB) cavities in the infected lungs from computed tomography (CT) scans. The authors' proposed method is based on a novel shape-based automated detection algorithm on CT scans followed by a fuzzy connectedness (FC) delineation procedure. In order to assess interaction between cavities and airways, the authors first roughly identified air-filled structures (airway, cavities, esophagus, etc.) by thresholding over Hounsfield unit of CT image. Then, airway and cavity structure detection was conducted within the support vector machine classification algorithm. Once airway and cavities were detected automatically, the authors extracted airway tree using a hybrid multiscale approach based on novel affinity relations within the FC framework and segmented cavities using intensity-based FC algorithm. At final step, the authors refined airway structures within the local regions of FC with finer control. Cavity segmentation results were compared to the reference truths provided by expert radiologists and cavity formation was tracked longitudinally from serial CT scans through shape and volume information automatically determined through the authors' proposed system. Morphological evolution of the cavitary TB were analyzed accordingly with this process. Finally, the authors computed the minimum distance between cavity surface and nearby airway structures by using the linear time distance transform algorithm to explore potential role of airways in cavity formation and morphological evolution. The proposed methodology was qualitatively and quantitatively evaluated on pulmonary CT images of rabbits experimentally infected with TB, and multiple markers such as cavity volume, cavity surface area, minimum distance from cavity surface to the nearest bronchial-tree, and longitudinal change of these markers (namely, morphological evolution of cavities) were determined precisely. While accuracy of the authors' cavity detection algorithm was 94.61%, airway detection part of the proposed methodology showed even higher performance by 99.8%. Dice similarity coefficients for cavitary segmentation experiments were found to be approximately 99.0% with respect to the reference truths provided by two expert radiologists (blinded to their evaluations). Moreover, the authors noted that volume derived from the authors' segmentation method was highly correlated with those provided by the expert radiologists (R(2) = 0.99757 and R(2) = 0.99496, p < 0.001, with respect to the observer 1 and observer 2) with an interobserver agreement of 98%. The authors quantitatively confirmed that cavity formation was positioned by the nearby bronchial-tree after exploring the respective spatial positions based on the minimum distance measurement. In terms of efficiency, the core algorithms take less than 2 min on a linux machine with 3.47 GHz CPU and 24 GB memory. The authors presented a fully automatic method for cavitary TB detection, quantification, and evaluation. The performance of every step of the algorithm was qualitatively and quantitatively assessed. With the proposed method, airways and cavities were automatically detected and subsequently delineated in high accuracy with heightened efficiency. Furthermore, not only morphological information of cavities were obtained through the authors' proposed framework, but their spatial relation to airways, and longitudinal analysis was also provided to get further insight on cavity formation in tuberculosis disease. To the authors' best of knowledge, this is the first study in computerized analysis of cavitary tuberculosis from CT scans.

NOVEL PREPROCESSING TECHNIQUE IN THE COMPUTER AIDED DETECTION OF BREAST CANCER

Analyzing medical images is the most challenging task in medical image processing. Computer Aided Detection (CAD) tool is the aid for the radiologists in analyzing such images for the effective detection and diagnosis of the disease. Such a CAD tool consists of Preprocessing, Segmentation and detection processes. In this study we have improved the preprocessing by using the Selective Median Filter (SMF) for the noise removal and modified the Local Range Modification (LRM) as modified LRM (MLRM) for the contrast enhancement to detect the breast cancer. We have tested the performance of some preprocessing methods and compared with the proposed method (MLRM). These methods (SMF and MLRM) had been tested for over 30 mammogram images and found the accuracy of 97.9% which is better than the other existing methods.

Computer-Aided Detection of Acute Pulmonary Embolism With 64-Slice Multi-Detector Row Computed Tomography

To evaluate the performance of a computer-aided detection (CAD) system for diagnosing peripheral acute pulmonary embolism (PE) with a 64-slice multi-detector row computed tomography (CT). Two radiologists investigated the accuracy of a software aimed at detecting peripheral clots (PECAD prototype, version 7; Siemens Medical Systems, Forchheim, Germany) by applying this tool for the analysis of the pulmonary arterial bed of 74 CT angiograms obtained with 64-slice dual-source CT (Definition; Siemens Medical Systems). These cases were retrospectively selected from a database of CT studies performed on the same CT unit, with a similar collimation (64 x 0.6 mm) and similar injection protocols. Patient selection was based on a variety of (1) scanning conditions, namely, nongated (n = 30), electrocardiography-gated (n = 30), and dual-energy CT angiograms (n = 14), and (2) image quality (IQ), namely, scans of excellent IQ (n = 53) and lower IQ due to lower levels of arterial enhancement and/or presence of noise (n = 21). The standard of truth was based on the 2 radiologists' consensus reading and the results of CAD. The software detected 80 of 93 peripheral clots present in the 21 patients (42 segmental and 38 subsegmental clots). The overall sensitivity (95% confidence interval) of the CAD tool was 86% (77%-92%) for detecting peripheral clots, 78% (64.5%-88%) at the segmental level and 97% (85.5%-99.9%) at the subsegmental level. Assuming normal vascular anatomy with 20 segmental and 40 subsegmental arteries, overall specificity and positive and negative predictive values (95% confidence interval) of the software were 91.8% (91%-92.6%), 18.4% (15%-22.4%), and 99.7% (99.5%-99.8%), respectively. A mean of 5.4 false positives was found per patient (total, 354 false positives), mainly linked to the presence of perivascular connective tissue (n = 119; 34%) and perivascular airspace consolidation (n = 97; 27%). The sensitivities (95% confidence interval) for the CAD tool were 91% (69.8%-99.3%) for dual-energy, 87% (59.3%-93.2%) for electrocardiography-gated, and 87% (73.5%-95.3%) for nongated scans (P > 0.05). No significant difference was found in the sensitivity of the CAD software when comparing the scans according to the scanning conditions and image quality. The evaluated CAD software has a good sensitivity in detecting peripheral PE, which is not influenced by the scanning conditions or the overall image quality.

Pulmonary Embolism: Computer-aided Detection at Multidetector Row Spiral Computed Tomography

We aimed to evaluate the feasibility and performance of a computer-aided detection (CAD) tool for automated detection of segmental and subsegmental pulmonary emboli. A CAD tool (ImageChecker CT, R2 Technology, Inc) for automated detection of pulmonary emboli was evaluated on multidetector-row CT studies of varying diagnostic quality in 23 patients (13 female, mean age 52 y) with pulmonary embolism (PE) and of 13 patients (all female, mean age 49 y) without PE. A collimation of 16 x 1 mm and a reconstructed section width of 1.25 mm had been used in each patient. The performance of the CAD tool for the detection of emboli in the segmental and subsegmental pulmonary arterial tree was assessed. Consensus reading of the same studies by 2 radiologists, with a third for adjudication, for the identification of segmental and subsegmental pulmonary emboli was used as the standard of reference. Consensus reading revealed 130 segmental pulmonary emboli and 107 subsegmental pulmonary emboli in the 23 patients with PE. All 23 patients with PE were correctly identified as having PE by the CAD system. In a vessel-by-vessel analysis, the sensitivity of the CAD algorithm was 92% (119/130) for the detection of segmental pulmonary emboli and 90% (92/107) for subsegmental pulmonary emboli. The overall specificity, positive predictive value (95% confidence interval) and negative predictive value (95% confidence interval) of the algorithm were 89.9%, 63.2% (57.9%-68.2%) and 97.7% (96.7%-98.4%), respectively. The average false positive rate of the CAD algorithm was 4.8 (range 1 to 9) false positive detection marks per case. CAD of segmental and subsegmental pulmonary emboli based on 1-mm multidetector-row CT studies is feasible. Application of CAD tools may improve the diagnostic accuracy and decrease the interpretation time of computed tomographic angiography for the detection of pulmonary emboli in the peripheral arterial tree and further enhance the acceptance of this test as the first line diagnostic modality for suspected PE.

Computer-assisted detection of pulmonary embolism: performance evaluation in consensus with experienced and inexperienced chest radiologists

The value of a computer-aided detection tool (CAD) as second reader in combination with experienced and inexperienced radiologists for the diagnosis of acute pulmonary embolism (PE) was assessed prospectively. Computed tomographic angiography (CTA) scans (64 x 0.6 mm collimation; 61.4 mm/rot table feed) of 56 patients (31 women, 34-89 years, mean = 66 years) with suspected PE were analysed by two experienced (R1, R2) and two inexperienced (R3, R4) radiologists for the presence and distribution of emboli using a five-point confidence rating, and by CAD. Informed consent was obtained from all patients. Results were compared with an independent reference standard. Inter-observer agreement was calculated by kappa, confidence assessed by ROC analysis. A total of 1,116 emboli [within mediastinal (n = 72), lobar (n = 133), segmental (n = 465) and subsegmental arteries (n = 455)] were included. CAD detected 343 emboli (sensitivity = 30.74%, correct-positive rate = 6.13/patient; false-positive rate = 4.1/patient). Inter-observer agreement was good (R1, R2: kappa = 0.84, 95% CI = 0.81-0.87; R3, R4: kappa = 0.79, 95% CI = 0.76-0.81). Extended inter-observer agreement was higher in mediastinal and lobar than in segmental and subsegmental arteries (kappa = 0.84-0.86 and kappa = 0.51-0.58 for mediastinal/lobar and segmental/subsegmental arteries, respectively P < 0.05). Agreement between experienced and inexperienced readers was improved by CAD (kappa = 0.60-0.62 and kappa = 0.69-0.72 before and after CAD consensus, respectively P < 0.05). The experienced outperformed the inexperienced readers (Az = 0.95, 0.93, 0.89 and 0.86 for R1-4, respectively, P < 0.05). CAD significantly improved overall performances of readers 3 and 4 (Az = 0.86 for R3, R4 and Az = 0.89 for R3, R4 with CAD, P < 0.05), by enhancing sensitivities in segmental/subsegmental arteries. CAD improved experienced readers' sensitivities in segmental/subsegmental arteries (sens. = 0.93 and 0.90 for R1, R2 before and 0.97 and 0.94 for R1, R2 after CAD consensus, P < 0.05), without significant improvement of their overall performances (P > 0.05). Particularly inexperienced readers benefit from consensus with CAD data, greatly improving detection of segmental and subsegmental emboli. This system is advocated as a second reader.

Computer-aided Classification of Interstitial Lung Diseases Via MDCT: 3D Adaptive Multiple Feature Method (3D AMFM)

Computer-aided detection algorithms applied to multidetector row CT (MDCT) lung image data sets have the potential to significantly alter clinical practice through the early, quantitative detection of pulmonary pathology. In this project, we have further developed a computer-aided detection tool, the adaptive multiple feature method (AMFM), for the detection of interstitial lung diseases based on MDCT-generated volumetric data. We performed MDCT (Siemens Sensation 16 or 64 120 kV, B50f convolution kernel, and <or=0.75-mm slice thickness) on 20 human volunteers recruited from four cohorts studied under an National Institutes of Health-sponsored Bioengineering Research Partnership Grant: 1) normal never smokers; 2) normal smokers; 3) those with emphysema, and 4) those with interstitial lung disease (total: 11 males, 9 females; age range 20-75 years, mean age 40 years). A total of 1,184 volumes of interest (VOIs; 21 x 21 pixels in plane) were marked by a senior radiologist and a senior pulmonologist as emphysema (EMPH, n = 287); ground-glass (GG, n = 147), honeycombing (HC, n = 137), normal nonsmokers (NN, n = 287), and normal smokers (NS, n = 326). For each VOI, we calculated 24 volumetric features, including statistical features (first-order features, run-length, and co-occurrence features), histogram, and fractal features. We compared two methods of classification (a Support Vector Machine (SVM) and a Bayesian classifier) using a 10-fold cross validation method and McNemar's test. The sensitivity of five patterns in the form of Bayesian/SVM was: EMPH: 91/93%; GG: 89/86%; HC: 93/90%; NN: 90/73%; and NS: 75/82%. The specificity of five patterns in the form of Bayesian/support vector machine was: EMPH: 98/98%; GG: 98/98%; HC: 99/99%; NN: 90/94%; and NS: 96/91%. We conclude that volumetric features including statistical features, histogram and fractal features can be successfully used in differentiation of parenchymal pathology associated with both emphysema and interstitial lung diseases. Additionally, support vector machine and Bayesian methods are comparable classifiers for characterization of interstitial lung diseases on MDCT images.

Computer-aided detection of therapy-induced leukoencephalopathy in pediatric acute lymphoblastic leukemia patients treated with intravenous high-dose methotrexate

The purpose of this study was to use objective quantitative magnetic resonance imaging (MRI) methods to develop a computer-aided detection (CAD) tool to differentiate white matter (WM) hyperintensities into either leukoencephalopathy (LE) induced by chemotherapy or normal maturational processes in children treated for acute lymphoblastic leukemia without irradiation. A combined MRI set consisting of T1-weighted, T2-weighted, proton-density-weighted and fluid-attenuated inversion recovery images and WM, gray matter and cerebrospinal fluid proportional volume maps from a spatially normalized atlas were analyzed with a neural network segmentation based on a Kohonen self-organizing map (SOM). Segmented maps were manually classified to identify the most hyperintense WM region and the normal-appearing genu region. Signal intensity differences normalized to the genu within each examination were generated for four time points in 228 children. A second Kohonen SOM was trained on the first examination data and divided the WM into normal-appearing or LE groups. Reviewing labels from the CAD tool revealed a consistency measure of 89.8% (167 of 186) within patients. The overall agreement between the CAD tool and the consensus reading of two trained observers was 84.1% (535 of 636), with 84.2% (170 of 202) agreement in the training set and 84.1% (365 of 434) agreement in the testing set. These results suggest that subtle therapy-induced LE can be objectively and reproducibly detected in children treated for cancer using this CAD approach based on relative differences in quantitative signal intensity measures normalized within each examination.

Computer-aided detection for mammography: improved algorithm performance with operator determined points characterized by new metrics

Recent prospective studies have suggested that user's confidence in computer-aided detection (CAD) marks and their tolerance of false positive (FP) marks play a role in the benefit gained from a CAD device. In this paper, we propose to (1) introduce new metrics to improve the characterization of the latest generation of CAD algorithms and (2) to describe the algorithm performance tradeoffs with user adjustable operating points. The performance of the CAD algorithm (V8.0, R2 Technology, Sunnyvale, CA) with user adjustable operating points was assessed on a multi-institutional database of 832 consecutive, biopsy-proven cancers detected by screening mammography and 345 clinically confirmed normal (four view) cases. The standard measurements of CAD performance were made—case sensitivity for microcalcifications and masses, and false positive (FP) marker rate on normal cases. In addition, new measurements of case specificity (the percentage of normal cases with no CAD marks) and two-view sensitivity (the percentage of cancer cases with a cancer visible ( n=628) and marked on both views) were used. It is likely that the case specificity measure reflects the impact of the CAD algorithm on the workflow of the radiologist. Further, a high value for the two-view sensitivity is likely to increase the radiologist's confidence in the CAD tool. This new CAD system allows the user to choose an operating point to best fit their clinical requirements.