Massive-training artificial neural network (MTANN) for reduction of false positives in computer-aided detection of polyps: Suppression of rectal tubes

One of the limitations of the current computer-aided detection (CAD) of polyps in CT colonography (CTC) is a relatively large number of false-positive (FP) detections. Rectal tubes (RTs) are one of the typical sources of FPs because a portion of a RT, especially a portion of a bulbous tip, often exhibits a cap-like shape that closely mimics the appearance of a small polyp. Radiologists can easily recognize and dismiss RT-induced FPs; thus, they may lose their confidence in CAD as an effective tool if the CAD scheme generates such "obvious" FPs due to RTs consistently. In addition, RT-induced FPs may distract radiologists from less common true positives in the rectum. Therefore, removal RT-induced FPs as well as other types of FPs is desirable while maintaining a high sensitivity in the detection of polyps. We developed a three-dimensional (3D) massive-training artificial neural network (MTANN) for distinction between polyps and RTs in 3D CTC volumetric data. The 3D MTANN is a supervised volume-processing technique which is trained with input CTC volumes and the corresponding "teaching" volumes. The teaching volume for a polyp contains a 3D Gaussian distribution, and that for a RT contains zeros for enhancement of polyps and suppression of RTs, respectively. For distinction between polyps and nonpolyps including RTs, a 3D scoring method based on a 3D Gaussian weighting function is applied to the output of the trained 3D MTANN. Our database consisted of CTC examinations of 73 patients, scanned in both supine and prone positions (146 CTC data sets in total), with optical colonoscopy as a reference standard for the presence of polyps. Fifteen patients had 28 polyps, 15 of which were 5-9 mm and 13 were 10-25 mm in size. These CTC cases were subjected to our previously reported CAD scheme that included centerline-based segmentation of the colon, shape-based detection of polyps, and reduction of FPs by use of a Bayesian neural network based on geometric and texture features. Application of this CAD scheme yielded 96.4% (27/28) by-polyp sensitivity with 3.1 (224/73) FPs per patient, among which 20 FPs were caused by RTs. To eliminate the FPs due to RTs and possibly other normal structures, we trained a 3D MTANN with ten representative polyps and ten RTs, and applied the trained 3D MTANN to the above CAD true- and false-positive detections. In the output volumes of the 3D MTANN, polyps were represented by distributions of bright voxels, whereas RTs and other normal structures partly similar to RTs appeared as darker voxels, indicating the ability of the 3D MTANN to suppress RTs as well as other normal structures effectively. Application of the 3D MTANN to the CAD detections showed that the 3D MTANN eliminated all RT-induced 20 FPs, as well as 53 FPs due to other causes, without removal of any true positives. Overall, the 3D MTANN was able to reduce the FP rate of the CAD scheme from 3.1 to 2.1 FPs per patient (33% reduction), while the original by-polyp sensitivity of 96.4% was maintained.     

Mixture of expert 3D massive-training ANNs for reduction of multiple types of false positives in CAD for detection of polyps in CT colonography

One of the major challenges in computer-aided detection (CAD) of polyps in CT colonography (CTC) is the reduction of false-positive detections (FPs) without a concomitant reduction in sensitivity. A large number of FPs is likely to confound the radiologist's task of image interpretation, lower the radiologist's efficiency, and cause radiologists to lose their confidence in CAD as a useful tool. Major sources of FPs generated by CAD schemes include haustral folds, residual stool, rectal tubes, the ileocecal valve, and extra-colonic structures such as the small bowel and stomach. Our purpose in this study was to develop a method for the removal of various types of FPs in CAD of polyps while maintaining a high sensitivity. To achieve this, we developed a "mixture of expert" three-dimensional (3D) massive-training artificial neural networks (MTANNs) consisting of four 3D MTANNs that were designed to differentiate between polyps and four categories of FPs: (1) rectal tubes, (2) stool with bubbles, (3) colonic walls with haustral folds, and (4) solid stool. Each expert 3D MTANN was trained with examples from a specific non-polyp category along with typical polyps. The four expert 3D MTANNs were combined with a mixing artificial neural network (ANN) such that different types of FPs could be removed. Our database consisted of 146 CTC datasets obtained from 73 patients whose colons were prepared by standard pre-colonoscopy cleansing. Each patient was scanned in both supine and prone positions. Radiologists established the locations of polyps through the use of optical-colonoscopy reports. Fifteen patients had 28 polyps, 15 of which were 5-9 mm and 13 were 10-25 mm in size. The CTC cases were subjected to our previously reported CAD method consisting of centerline-based extraction of the colon, shape-based detection of polyp candidates, and a Bayesian-ANN-based classification of polyps. The original CAD method yielded 96.4% (27/28) by-polyp sensitivity with an average of 3.1 (224/73) FPs per patient. The mixture of expert 3D MTANNs removed 63% (142/224) of the FPs without the loss of any true positive; thus, the FP rate of our CAD scheme was improved to 1.1 (82/73) FPs per patient while the original sensitivity was maintained. By use of the mixture of expert 3D MTANNs, the specificity of a CAD scheme for detection of polyps in CTC was substantially improved while a high sensitivity was maintained.     

Computerized detection of colorectal masses in CT colonography based on fuzzy merging and wall-thickening analysis

In recent years, several computer-aided detection (CAD) schemes have been developed for the detection of polyps in CT colonography (CTC). However, few studies have addressed the problem of computerized detection of colorectal masses in CTC. This is mostly because masses are considered to be well visualized by a radiologist because of their size and invasiveness. Nevertheless, the automated detection of masses would naturally complement the automated detection of polyps in CTC and would produce a more comprehensive computer aid to radiologists. Therefore, in this study, we identified some of the problems involved with the computerized detection of masses, and we developed a scheme for the computerized detection of masses that can be integrated into a CAD scheme for the detection of polyps. The performance of the mass detection scheme was evaluated by application to clinical CTC data sets. CTC was performed on 82 patients with helical CT scanners and reconstruction intervals of 1.0-5.0 mm in the supine and prone positions. Fourteen patients (17%) had a total of 14 masses of 30-50 mm, and sixteen patients (20%) had a total of 30 polyps 5-25 mm in diameter. Four patients had both polyps and masses. Fifty-six of the patients (68%) were normal. The CTC data were interpolated linearly to yield isotropic data sets, and the colon was extracted by use of a knowledge-guided segmentation technique. Two methods, fuzzy merging and wall-thickening analysis, were developed for the detection of masses. The fuzzy merging method detected masses with a significant intraluminal component by separating the initial CAD detections of locally cap-like shapes within the colonic wall into mass candidates and polyp candidates. The wall-thickening analysis detected nonintraluminal masses by searching the colonic wall for abnormal thickening. The final regions of the mass candidates were extracted by use of a level set method based on a fast marching algorithm. False-positive (FP) detections were reduced by a quadratic discriminant classifier. The performance of the scheme was evaluated by use of a leave-one-out (round-robin) method with by-patient elimination. All but one of the 14 masses, which was partially cut off from the CTC data set in both supine and prone positions, were detected. The fuzzy merging method detected 11 of the masses, and the wall-thickening analysis detected 3 of the masses including all nonintraluminal masses. In combination, the two methods detected 13 of the 14 masses with 0.21 FPs per patient on average based on the leave-one-out evaluation. Most FPs were generated by extrinsic compression of the colonic wall that would be recognized easily and quickly by a radiologist. The mass detection methods did not affect the result of the polyp detection. The results indicate that the scheme is potentially useful in providing a high-performance CAD scheme for the detection of colorectal neoplasms in CTC.     

Reduction of false positives on the rectal tube in computer-aided detection for CT colonography

PURPOSE: To eliminate false-positive (FP) polyp detections on the rectal tube (RT) in CT colonography (CTC) computer-aided detection (CAD). METHODS: We use a three-stage approach to detect the RT: detect the RT shaft, track the tube to the tip and label all the voxels that belong to the RT. We applied our RT detection algorithm on a CTC dataset consisting of 80 datasets (40 patients scanned in both prone and supine positions). Two different types of RTs were present, characterized by differences in shaft/bulb diameters, wall intensities, and shape of tip. RESULTS: The algorithm detected 90% of RT shafts and completely tracked 72% of them. We labeled all the voxels belonging to the completely tracked RTs (72%) and in 11 out of 80 (14%) cases the RT voxels were partially labeled. We obtained a 9.2% reduction of the FPs in the initial polyp candidates' population, and a 7.9% reduction of the FPs generated by our CAD system. None of the true-positive detections were mislabeled. CONCLUSIONS: The algorithm detects the RTs with good accuracy, is robust with respect to the two different types of RT used in our study, and is effective at reducing the number of RT FPs reported by our CAD system.     

Wavelet method for CT colonography computer-aided polyp detection

Computed tomographic colonography (CTC) computer aided detection (CAD) is a new method to detect colon polyps. Colonic polyps are abnormal growths that may become cancerous. Detection and removal of colonic polyps, particularly larger ones, has been shown to reduce the incidence of colorectal cancer. While high sensitivities and low false positive rates are consistently achieved for the detection of polyps sized 1 cm or larger, lower sensitivities and higher false positive rates occur when the goal of CAD is to identify "medium"-sized polyps, 6-9 mm in diameter. Such medium-sized polyps may be important for clinical patient management. We have developed a wavelet-based postprocessor to reduce false positives for this polyp size range. We applied the wavelet-based postprocessor to CTC CAD findings from 44 patients in whom 45 polyps with sizes of 6-9 mm were found at segmentally unblinded optical colonoscopy and visible on retrospective review of the CT colonography images. Prior to the application of the wavelet-based postprocessor, the CTC CAD system detected 33 of the polyps (sensitivity 73.33%) with 12.4 false positives per patient, a sensitivity comparable to that of expert radiologists. Fourfold cross validation with 5000 bootstraps showed that the wavelet-based postprocessor could reduce the false positives by 56.61% (p <0.001), to 5.38 per patient (95% confidence interval [4.41, 6.34]), without significant sensitivity degradation (32/45, 71.11%, 95% confidence interval [66.39%, 75.74%], p=0.1713). We conclude that this wavelet-based postprocessor can substantially reduce the false positive rate of our CTC CAD for this important polyp size range.     

Feature-guided analysis for reduction of false positives in CAD of polyps for computed tomographic colonography

We evaluated the effect of our novel technique of feature-guided analysis of polyps on the reduction of false-positive (FP) findings generated by our computer-aided diagnosis (CAD) scheme for the detection of polyps from computed tomography colonographic data sets. The detection performance obtained by use of feature-guided analysis in the segmentation and feature analysis of polyp candidates was compared with that obtained by use of our previously employed fuzzy clustering technique. We also evaluated the effect of a feature called modified gradient concentration (MGC) on the detection performance. A total of 144 data sets, representing prone and supine views of 72 patients that included 14 patients with 21 colorectal polyps 5-25 mm in diameter, were used in the evaluation. At a 100% by-patient (95% by-polyp) detection sensitivity, the FP rate of our CAD scheme with feature-guided analysis based on round-robin evaluation was 1.3 (1.5) FP detections per patient. This corresponds to a 70-75% reduction in the number of FPs obtained by use of fuzzy clustering at the same sensitivity levels. Application of the MGC feature instead of our previously used gradient concentration feature did not improve the detection result. The results indicate that feature-guided analysis is useful for achieving high sensitivity and a low FP rate in our CAD scheme.     

Registration of central paths and colonic polyps between supine and prone scans in computed tomography colonography: pilot study

Computed tomography colonography (CTC) is a minimally invasive method that allows the evaluation of the colon wall from CT sections of the abdomen/pelvis. The primary goal of CTC is to detect colonic polyps, precursors to colorectal cancer. Because imperfect cleansing and distension can cause portions of the colon wall to be collapsed, covered with water, and/or covered with retained stool, patients are scanned in both prone and supine positions. We believe that both reading efficiency and computer aided detection (CAD) of CTC images can be improved by accurate registration of data from the supine and prone positions. We developed a two-stage approach that first registers the colonic central paths using a heuristic and automated algorithm and then matches polyps or polyp candidates (CAD hits) by a statistical approach. We evaluated the registration algorithm on 24 patient cases. After path registration, the mean misalignment distance between prone and supine identical anatomic landmarks was reduced from 47.08 to 12.66 mm, a 73% improvement. The polyp registration algorithm was specifically evaluated using eight patient cases for which radiologists identified polyps separately for both supine and prone data sets, and then manually registered corresponding pairs. The algorithm correctly matched 78% of these pairs without user input. The algorithm was also applied to the 30 highest-scoring CAD hits in the prone and supine scans and showed a success rate of 50% in automatically registering corresponding polyp pairs. Finally, we computed the average number of CAD hits that need to be manually compared in order to find the correct matches among the top 30 CAD hits. With polyp registration, the average number of comparisons was 1.78 per polyp, as opposed to 4.28 comparisons without polyp registration.     

CT colonography: does improved z resolution help computer-aided polyp detection?

Multislice helical CT offers several retrospective choices of longitudinal (z) resolution at a given detector collimation setting. We sought to determine the effect of z resolution on the performance of a computer-aided colonic polyp detector, since a human reader and a computer-aided polyp detector may have optimal performances at different z resolutions. We ran a computer-aided polyp detection algorithm on phantom data sets as well as data obtained from a single patient. All data were reconstructed at various slice thicknesses ranging from 1.25 to 10 mm. We studied the performance of the detector at various ranges of polyp sizes using free-response receiver-operating characteristic analyses. We also studied contrast-to-noise ratios (CNR) as a function of slice thickness and polyp size. For the phantom data, reducing the slice thickness from 5 to 1.25 mm improves sensitivity from 84.5% to 98.3% (all polyps), from 61.4% to 95.5% (polyps in the range [0, 5) mm) and from 97.7% to 100% (polyps in the range [5, 10) mm) at a false positive rate of 20 per data set. For polyps larger than 10 mm, there is no significant improvement in detection sensitivity when slice thickness is reduced. CNRs showed expected behavior with slice thickness and polyp size, but in all cases remained high (> 4). The results for the patient data followed similar patterns to that of the phantom case. Thus we conclude that for this detector, the optimal slice thickness is dependent upon the size of the smallest polyps to be detected. For detection of polyps 10 mm and larger, reconstruction of 5 mm sections may be sufficient. Further study is required to generalize these results to a broader population of patients scanned on different scanners.     

Automatic polyp detection and measurement with computed tomographic colonography: A phantom study

Purpose

The purpose of this study is to assess the performance of computer-aided detection (CAD) software in detecting and measuring polyps for CT Colonography, based on an in vitro phantom study.

Material and methods

A colon phantom was constructed with a PVC pipe of 3.8 cm diameter. Nine simulated polyps of various sizes (3.2mm-25.4mm) were affixed inside the phantom that was placed in a water bath. The phantom was scanned on a 64-slice CT scanner with tube voltage of 120 kV and current of 205 mAs. Two separate scans were performed, with different slice thickness and reconstruction interval. The first scan (thin) had a slice thickness of 1mm and reconstruction interval 0.5mm. The second scan (thick) had a slice thickness of 2mm and reconstruction interval of 1mm. Images from both scans were processed using CT Colonography software that automatically segments the colon phantom and applies CAD that automatically highlights and provides the size (maximum and minimum diameters, volume) of each polyp. Two readers independently measured each polyp (two orthogonal diameters) using both 2D and 3D views. Readers’ manual measurements (diameters) and automatic measurements from CAD (diameters and volume) were compared to actual polyp sizes as measured by mechanical calipers.

Results

All polyps except the smallest (3.2mm) were detected by CAD. CAD achieved 100% sensitivity in detecting polyps ≥6mm. Mean errors in CAD automated volume measurements for thin and thick slice scans were 8.7% and 6.8%, respectively. Almost all CAD and manual readers’ 3D measurements overestimated the size of polyps to variable extent. Both over- and underestimation of polyp sizes were observed in the readers’ manual 2D measurements. Overall, Reader 1 (expert) had smaller mean error than Reader 2 (non-expert).

Conclusion

CAD provided accurate size measurements for all polyps, and results were comparable to the two readers'' manual measurements     

Eigenspace Template Matching for Detection of Lacunar Infarcts on MR Images

Detection of lacunar infarcts is important because their presence indicates an increased risk of severe cerebral infarction. However, accurate identification is often hindered by the difficulty in distinguishing between lacunar infarcts and enlarged Virchow-Robin spaces. Therefore, we developed a computer-aided detection (CAD) scheme for the detection of lacunar infarcts. Although our previous CAD method indicated a sensitivity of 96.8 % with 0.71 false positives (FPs) per slice, further reduction of FPs remained an issue for the clinical application. Thus, the purpose of this study is to improve our CAD scheme by using template matching in the eigenspace. Conventional template matching is useful for the reduction of FPs, but it has the following two pitfalls: (1) It needs to maintain a large number of templates to improve the detection performance, and (2) calculation of the cross-correlation coefficient with these templates is time consuming. To solve these problems, we used template matching in the lower dimension space made by a principal component analysis. Our database comprised 1,143 T₁- and T₂-weighted images obtained from 132 patients. The proposed method was evaluated by using twofold cross-validation. By using this method, 34.1 % of FPs was eliminated compared with our previous method. The final performance indicated that the sensitivity of the detection of lacunar infarcts was 96.8 % with 0.47 FPs per slice. Therefore, the modified CAD scheme could improve FP rate without a significant reduction in the true positive rate.     

Computer aided detection of clusters of microcalcifications on full field digital mammograms

We are developing a computer-aided detection (CAD) system to identify microcalcification clusters (MCCs) automatically on full field digital mammograms (FFDMs). The CAD system includes six stages: preprocessing; image enhancement; segmentation of microcalcification candidates; false positive (FP) reduction for individual microcalcifications; regional clustering; and FP reduction for clustered microcalcifications. At the stage of FP reduction for individual microcalcifications, a truncated sum-of-squares error function was used to improve the efficiency and robustness of the training of an artificial neural network in our CAD system for FFDMs. At the stage of FP reduction for clustered microcalcifications, morphological features and features derived from the artificial neural network outputs were extracted from each cluster. Stepwise linear discriminant analysis (LDA) was used to select the features. An LDA classifier was then used to differentiate clustered microcalcifications from FPs. A data set of 96 cases with 192 images was collected at the University of Michigan. This data set contained 96 MCCs, of which 28 clusters were proven by biopsy to be malignant and 68 were proven to be benign. The data set was separated into two independent data sets for training and testing of the CAD system in a cross-validation scheme. When one data set was used to train and validate the convolution neural network (CNN) in our CAD system, the other data set was used to evaluate the detection performance. With the use of a truncated error metric, the training of CNN could be accelerated and the classification performance was improved. The CNN in combination with an LDA classifier could substantially reduce FPs with a small tradeoff in sensitivity. By using the free-response receiver operating characteristic methodology, it was found that our CAD system can achieve a cluster-based sensitivity of 70, 80, and 90 % at 0.21, 0.61, and 1.49 FPs/image, respectively. For case-based performance evaluation, a sensitivity of 70, 80, and 90 % can be achieved at 0.07, 0.17, and 0.65 FPs/image, respectively. We also used a data set of 216 mammograms negative for clustered microcalcifications to further estimate the FP rate of our CAD system. The corresponding FP rates were 0.15, 0.31, and 0.86 FPs/image for cluster-based detection when negative mammograms were used for estimation of FP rates.     

基于同构化改进的U-Net结直肠息肉分割方法

结肠镜检查广泛应用于结直肠癌的早期筛查和诊疗,但仅靠人工判读结肠息肉漏检率较高,有研究统计可达25％.基于深度学习的计算机辅助技术有助于提高息肉检测率,但目前深度学习的主流分割网络U-Net存在着两个局限:一是编解码的输出特征图之间存在着语义鸿沟;二是U-Net的双层卷积单元无法学习多尺度信息;割裂地看待容易使模型陷入...     

基于尺寸自适应深度神经网络的胸部CT图像肺结节检测

使用计算机断层扫描(CT)筛查肺结节是早期肺癌诊断的重要手段.但由于肺结节在形状、大小和位置上有存在很大的差异,目前肺结节尤其是小结节的自动检测依然具有挑战性.为了实现高灵敏度的肺结节检测,提出一种新的计算机辅助检测系统,该系统采用两种新的策略:尺寸自适应候选检测(SACD)和尺寸自适应假阳性抑制(SAFPR).首先,...     

边缘先验信息下的多类型肠道息肉图像分类网络

肠道息肉的分级能够为内窥镜医生提供辅助诊断,对需要及时处理的高风险息肉和可以暂缓处理的低风险息肉进行区分。现有的基于深度学习息肉分类算法不能很好地区分类间相似性高的图像,针对息肉分级任务有待改进。提出一个包含边缘检测阶段、边缘特征描述提取阶段以及息肉分类阶段的边缘先验信息下的多类型肠道息肉图像分类网络。首先,在边缘检测阶段的跳跃连接层处,设计并嵌入反向注意力边缘监督模块以更好地捕获息肉边缘细节信息;其次,在内窥镜医生先验知识的指导下分别通过统计息肉边缘像素点个数和凹凸性来表示息肉边缘周长大小和光滑性特征,以此来补充神经网络特征提取的不足;最后,在分类网络的DenseBlock4后加入通道注意力自适应地捕获判别性特征。所构建的数据集来自丽水市人民医院消化内镜中心2018年至2019年的脱敏数据,样本量含1 050幅原始图像。在构建的四分类数据集上进行五折交叉验证,达到了77.29%的总体准确率,相比于已有算法的最好结果提高了6.46%。融合边缘先验信息的分类网络能够有效地对非腺瘤性息肉与低级别腺瘤性息肉、高级别腺瘤性息肉与腺癌这两组类间高相似度的息肉图像进行区分,增加网络的鲁棒性并提高网络的分类性能,在有限的训练数据集下为医生诊断提供辅助意见。     

Improved classifier for computer-aided polyp detection in CT colonography by nonlinear dimensionality reduction

Computer-aided detection (CAD) has been shown to be feasible for polyp detection on computed tomography (CT) scans. After initial detection, the dataset of colonic polyp candidates has large-scale and high dimensional characteristics. In this article, we propose a nonlinear dimensionality reduction method based on diffusion map and locally linear embedding (DMLLE) for large-scale datasets. By selecting partial data as landmarks, we first map these points into a low dimensional embedding space using the diffusion map. The embedded landmarks can be viewed as a skeleton of whole data in the low dimensional space. Then by using the locally linear embedding algorithm, nonlandmark samples are mapped into the same low dimensional space according to their nearest landmark samples. The local geometry is preserved in both the original high dimensional space and the embedding space. In addition, DMLLE provides a faithful representation of the original high dimensional data at coarse and fine scales. Thus, it can capture the intrinsic distance relationship between samples and reduce the influence of noisy features, two aspects that are crucial to achieving high classifier performance. We applied the proposed DMLLE method to a colonic polyp dataset of 175 269 polyp candidates with 155 features. Visual inspection shows that true polyps with similar shapes are mapped to close vicinity in the low dimensional space. We compared the performance of a support vector machine (SVM) classifier in the low dimensional embedding space with that in the original high dimensional space, SVM with principal component analysis dimensionality reduction and SVM committee using feature selection technology. Free-response receiver operating characteristic analysis shows that by using our DMLLE dimensionality reduction method, SVM achieves higher sensitivity with a lower false positive rate compared with other methods. For 6-9 mm polyps (193 true polyps contained in test set), when the number of false positives per patient is 9, SVM with DMLLE improves the average sensitivity from 70% to 83% compared with that of an SVM committee classifier which is a state-of-the-art method for colonic polyp detection (p<0.001).     

无线胶囊内窥镜图像小肠息肉的自动识别

息肉是小肠常见疾病之一。无线胶囊内窥镜(WCE)是检查小肠疾病的常规手段,但每次检查都会产生大量图像,却仅可能包含少数病变图像。目前WCE病变的筛查高度依赖于医生的临床经验,耗时费力,且可能发生漏检或误检,因此实现WCE图像小肠息肉的自动识别意义重大。基于深度学习框架,结合数据增强技术和迁移学习策略,提出实现小肠息肉识别的新方法。基于原始数据集(包含4 300张正常图像和429张息肉图像)和拓展数据集(包含6 920张正常图像和6 864张息肉图像),对比分析不同的深度卷积神经网络模型(AlexNet、VGGNet和GoogLeNet)对息肉的识别效果。实验结果表明,在随机初始化的卷积神经网络中,GoogLeNet模型对息肉的识别效果最好,其敏感性、特异性和准确性分别达到97.18%、98.78%和97.99%,说明增加网络深度可以有效提高识别率。但网络深度增加需要更高的硬件配置和训练时间,因此结合迁移学习策略,AlexNet模型的敏感性、特异性和准确性分别达到了96.57%、98.89%和97.74%,AUC为0.996,表明该方法能有效提高模型整体性能,同时降低对训练时间和实验配置的要求。与传统手工提取图像特征或仅基于深层卷积神经网络进行分类的方法相比,所提出的方法可以在有限的训练数据和实验环境下为小肠息肉的自动识别提供有效的解决方案,有望帮助医生高效完成基于WCE检查的消化道疾病的精准诊断。     

Computer-aided detection of early interstitial lung diseases using low-dose CT images

This study aims to develop a new computer-aided detection (CAD) scheme to detect early interstitial lung disease (ILD) using low-dose computed tomography (CT) examinations. The CAD scheme classifies each pixel depicted on the segmented lung areas into positive or negative groups for ILD using a mesh-grid-based region growth method and a multi-feature-based artificial neural network (ANN). A genetic algorithm was applied to select optimal image features and the ANN structure. In testing each CT examination, only pixels selected by the mesh-grid region growth method were analyzed and classified by the ANN to improve computational efficiency. All unselected pixels were classified as negative for ILD. After classifying all pixels into the positive and negative groups, CAD computed a detection score based on the ratio of the number of positive pixels to all pixels in the segmented lung areas, which indicates the likelihood of the test case being positive for ILD. When applying to an independent testing dataset of 15 positive and 15 negative cases, the CAD scheme yielded the area under receiver operating characteristic curve (AUC = 0.884 ± 0.064) and 80.0% sensitivity at 85.7% specificity. The results demonstrated the feasibility of applying the CAD scheme to automatically detect early ILD using low-dose CT examinations.     

A method to test the reproducibility and to improve performance of computer-aided detection schemes for digitized mammograms

The purpose of this study is to develop a new method for assessment of the reproducibility of computer-aided detection (CAD) schemes for digitized mammograms and to evaluate the possibility of using the implemented approach for improving CAD performance. Two thousand digitized mammograms (representing 500 cases) with 300 depicted verified masses were selected in the study. Series of images were generated for each digitized image by resampling after a series of slight image rotations. A CAD scheme developed in our laboratory was applied to all images to detect suspicious mass regions. We evaluated the reproducibility of the scheme using the detection sensitivity and false-positive rates for the original and resampled images. We also explored the possibility of improving CAD performance using three methods of combining results from the original and resampled images, including simple grouping, averaging output scores, and averaging output scores after grouping. The CAD scheme generated a detection score (from 0 to 1) for each identified suspicious region. A region with a detection score >0.5 was considered as positive. The CAD scheme detected 238 masses (79.3% case-based sensitivity) and identified 1093 false-positive regions (average 0.55 per image) in the original image dataset. In eleven repeated tests using original and ten sets of rotated and resampled images, the scheme detected a maximum of 271 masses and identified as many as 2359 false-positive regions. Two hundred and eighteen masses (80.4%) and 618 false-positive regions (26.2%) were detected in all 11 sets of images. Combining detection results improved reproducibility and the overall CAD performance. In the range of an average false-positive detection rate between 0.5 and 1 per image, the sensitivity of the scheme could be increased approximately 5% after averaging the scores of the regions detected in at least four images. At low false-positive rate (e.g., < or =average 0.3 per image), the grouping method alone could increase CAD sensitivity by 7%. The study demonstrated that reproducibility of a CAD scheme can be tested using a set of slightly rotated and resampled images. Because the reproducibility of true-positive detections is generally higher than that of false-positive detections, combining detection results generated from subsets of rotated and resampled images could improve both reproducibility and overall performance of CAD schemes.     

Bilateral analysis based false positive reduction for computer-aided mass detection

We have developed a false positive (FP) reduction method based on analysis of bilateral mammograms for computerized mass detection systems. The mass candidates on each view were first detected by our unilateral computer-aided detection (CAD) system. For each detected object, a regional registration technique was used to define a region of interest (ROI) that is "symmetrical" to the object location on the contralateral mammogram. Texture features derived from the spatial gray level dependence matrices and morphological features were extracted from the ROI containing the detected object on a mammogram and its corresponding ROI on the contralateral mammogram. Bilateral features were then generated from corresponding pairs of unilateral features for each object. Two linear discriminant analysis (LDA) classifiers were trained from the unilateral and the bilateral feature spaces, respectively. Finally, the scores from the unilateral LDA classifier and the bilateral LDA asymmetry classifier were fused with a third LDA whose output score was used to distinguish true mass from FPs. A data set of 341 cases of bilateral two-view mammograms was used in this study, of which 276 cases with 552 bilateral pairs contained 110 malignant and 166 benign biopsy-proven masses and 65 cases with 130 bilateral pairs were normal. The mass data set was divided into two subsets for twofold cross-validation training and testing. The normal data set was used for estimation of FP rates. It was found that our bilateral CAD system achieved a case-based sensitivity of 70%, 80%, and 85% at average FP rates of 0.35, 0.75, and 0.95 FPs/image, respectively, on the test data sets with malignant masses. In comparison to the average FP rates for the unilateral CAD system of 0.58, 1.33, and 1.63, respectively, at the corresponding sensitivities, the FP rates were reduced by 40%, 44%, and 42% with the bilateral symmetry information. The improvement was statistically significance (p < 0.05) as estimated by JAFROC analysis.