Search

Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Purpose: The purpose of this study was to develop and evaluate lung cancer segmentation with a pretrained model and transfer learning. The pretrained model was constructed from an artificial dataset generated using a generative adversarial network (GAN). Materials and Methods: Three public datasets containing images of lung nodules/lung cancers were used: LUNA16 dataset, Decathlon lung dataset, and NSCLC radiogenomics. The LUNA16 dataset was used to generate an artificial dataset for lung cancer segmentation with the help of the GAN and 3D graph cut. Pretrained models were then constructed from the artificial dataset. Subsequently, the main segmentation model was constructed from the pretrained models and the Decathlon lung dataset. Finally, the NSCLC radiogenomics dataset was used to evaluate the main segmentation model. The Dice similarity coefficient (DSC) was used as a metric to evaluate the segmentation performance. Results: The mean DSC for the NSCLC radiogenomics dataset improved overall when using the pretrained models. At maximum, the mean DSC was 0.09 higher with the pretrained model than that without it. Conclusion: The proposed method comprising an artificial dataset and a pretrained model can improve lung cancer segmentation as confirmed in terms of the DSC metric. Moreover, the construction of the artificial dataset for the segmentation using the GAN and 3D graph cut was found to be feasible., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2021 Nishio, Fujimoto, Matsuo, Muramatsu, Sakamoto and Fujita.)

Objective: The first aim of this study was to determine the performance of a deep learning object detection technique in the detection of maxillary sinuses on panoramic radiographs. The second aim was to clarify the performance in the classification of maxillary sinus lesions compared with healthy maxillary sinuses., Methods: The imaging data for healthy maxillary sinuses (587 sinuses, Class 0), inflamed maxillary sinuses (416 sinuses, Class 1), cysts of maxillary sinus regions (171 sinuses, Class 2) were assigned to training, testing 1, and testing 2 data sets. A learning process of 1000 epochs with the training images and labels was performed using DetectNet, and a learning model was created. The testing 1 and testing 2 images were applied to the model, and the detection sensitivities and the false-positive rates per image were calculated. The accuracies, sensitivities and specificities were determined for distinguishing the inflammation group (Class 1) and cyst group (Class 2) with respect to the healthy group (Class 0)., Results: Detection sensitivities of healthy (Class 0) and inflamed (Class 1) maxillary sinuses were 100% for both testing 1 and testing 2 data sets, whereas they were 98 and 89% for cysts of the maxillary sinus regions (Class 2). False-positive rates per image were nearly 0.00. Accuracies, sensitivities and specificities for diagnosis maxillary sinusitis were 90-91%, 88-85%, and 91-96%, respectively; for cysts of the maxillary sinus regions, these values were 97-100%, 80-100%, and 100-100%, respectively., Conclusion: Deep learning could reliably detect the maxillary sinuses and identify maxillary sinusitis and cysts of the maxillary sinus regions., Advances in Knowledge: This study using a deep leaning object detection technique indicated that the detection sensitivities of maxillary sinuses were high and the performance of maxillary sinus lesion identification was ≧80%. In particular, performance of sinusitis identification was ≧90%.

Objectives: Dental state plays an important role in forensic radiology in case of large scale disasters. However, dental information stored in dental clinics are not standardized or electronically filed in general. The purpose of this study is to develop a computerized system to detect and classify teeth in dental panoramic radiographs for automatic structured filing of the dental charts. It can also be used as a preprocessing step for computerized image analysis of dental diseases., Methods: One hundred dental panoramic radiographs were employed for training and testing an object detection network using fourfold cross-validation method. The detected bounding boxes were then classified into four tooth types, including incisors, canines, premolars, and molars, and three tooth conditions, including nonmetal restored, partially restored, and completely restored, using classification network. Based on the visualization result, multisized image data were used for the double input layers of a convolutional neural network. The result was evaluated by the detection sensitivity, the number of false-positive detection, and classification accuracies., Results: The tooth detection sensitivity was 96.4% with 0.5 false positives per case. The classification accuracies for tooth types and tooth conditions were 93.2% and 98.0%. Using the double input layer network, 6 point increase in classification accuracy was achieved for the tooth types., Conclusions: The proposed method can be useful in automatic filing of dental charts for forensic identification and preprocessing of dental disease prescreening purposes.

Purpose: To develop and evaluate a three-dimensional (3D) generative model of computed tomography (CT) images of lung nodules using a generative adversarial network (GAN). To guide the GAN, lung nodule size was used., Materials and Methods: A public CT dataset of lung nodules was used, from where 1182 lung nodules were obtained. Our proposed GAN model used masked 3D CT images and nodule size information to generate images. To evaluate the generated CT images, two radiologists visually evaluated whether the CT images with lung nodule were true or generated, and the diagnostic ability was evaluated using receiver-operating characteristic analysis and area under the curves (AUC). Then, two models for classifying nodule size into five categories were trained, one using the true and the other using the generated CT images of lung nodules. Using true CT images, the classification accuracy of the sizes of the true lung nodules was calculated for the two classification models., Results: The sensitivity, specificity, and AUC of the two radiologists were respectively as follows: radiologist 1: 81.3%, 37.7%, and 0.592; radiologist 2: 77.1%, 30.2%, and 0.597. For categorization of nodule size, the mean accuracy of the classification model constructed with true CT images was 85% (range 83.2-86.1%), and that with generated CT images was 85% (range 82.2-88.1%)., Conclusions: Our results show that it was possible to generate 3D CT images of lung nodules that could be used to construct a classification model of lung nodule size without true CT images., (Copyright © 2020 Elsevier Ltd. All rights reserved.)

Objective: The aim of this study was to compare time and storage space requirements, diagnostic performance, and consistency among 3 image recognition convolutional neural networks (CNNs) in the evaluation of the relationships between the mandibular third molar and the mandibular canal on panoramic radiographs., Study Design: Of 600 panoramic radiographs, 300 each were assigned to noncontact and contact groups based on the relationship between the mandibular third molar and the mandibular canal. The CNNs were trained twice by using cropped image patches with sizes of 70 × 70 pixels and 140 × 140 pixels. Time and storage space were measured for each system. Accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC) were determined. Intra-CNN and inter-CNN consistency values were calculated., Results: Time and storage space requirements depended on the depth of CNN layers and number of learned parameters, respectively. The highest AUC values ranged from 0.88 to 0.93 in the CNNs created by 70 × 70 pixel patches, but there were no significant differences in diagnostic performance among any of the models with smaller patches. Intra-CNN and inter-CNN consistency values were good or very good for all CNNs., Conclusions: The size of the image patches should be carefully determined to ensure acquisition of high diagnostic performance and consistency., (Copyright © 2020. Published by Elsevier Inc.)

Training of a convolutional neural network (CNN) generally requires a large dataset. However, it is not easy to collect a large medical image dataset. The purpose of this study is to investigate the utility of synthetic images in training CNNs and to demonstrate the applicability of unrelated images by domain transformation. Mammograms showing 202 benign and 212 malignant masses were used for evaluation. To create synthetic data, a cycle generative adversarial network was trained with 599 lung nodules in computed tomography (CT) and 1430 breast masses on digitized mammograms (DDSM). A CNN was trained for classification between benign and malignant masses. The classification performance was compared between the networks trained with the original data, augmented data, synthetic data, DDSM images, and natural images (ImageNet dataset). The results were evaluated in terms of the classification accuracy and the area under the receiver operating characteristic curves (AUC). The classification accuracy improved from 65.7% to 67.1% with data augmentation. The use of an ImageNet pretrained model was useful (79.2%). Performance was slightly improved when synthetic images or the DDSM images only were used for pretraining (67.6 and 72.5%, respectively). When the ImageNet pretrained model was trained with the synthetic images, the classification performance slightly improved (81.4%), although the difference in AUCs was not statistically significant. The use of the synthetic images had an effect similar to the DDSM images. The results of the proposed study indicated that the synthetic data generated from unrelated lesions by domain transformation could be used to increase the training samples., Competing Interests: Declaration of competing interest None declared., (Copyright © 2020 Elsevier Ltd. All rights reserved.)

Early detection of glaucoma is important to slow down progression of the disease and to prevent total vision loss. Retinal fundus photography is frequently obtained for various eye disease diagnosis and record and is a suitable screening exam for its simplicity and low cost. However, the number of ophthalmologists who are specialized in glaucoma diagnosis is limited. We have been studying automated schemes for detection of nerve fiber layer defects and analysis of optic disc deformation, two major signs of glaucoma, in assisting ophthalmologists' accurate and efficient diagnosis. In this chapter, our recent progress in computerized methods is discussed.

Objective: The aim of this study was to investigate whether a deep learning object detection technique can automatically detect and classify radiolucent lesions in the mandible on panoramic radiographs., Study Design: Panoramic radiographs of patients with mandibular radiolucent lesions of 10 mm or greater, including ameloblastomas, odontogenic keratocysts, dentigerous cysts, radicular cysts, and simple bone cysts, were included. Lesion labels, including region of interest coordinates, were created in text format. In total, 210 training images and labels were imported into the deep learning GPU training system (DIGITS). A learning model was created using the deep neural network DetectNet. Two testing data sets (testing 1 and 2) were applied to the learning model. Similarities and differences between the prediction and ground-truth images were evaluated using Intersection over Union (IoU). Sensitivity and false-positive rate per image were calculated using an IoU threshold of 0.6. The detection performance for each disease was assessed using multiclass learning., Results: Sensitivity was 0.88 for both testing 1 and 2. The false-positive rate per image was 0.00 for testing 1 and 0.04 for testing 2. The best combination of detection and classification sensitivity occurred with dentigerous cysts., Conclusions: Radiolucent lesions of the mandible can be detected with high sensitivity using deep learning., (Copyright © 2019. Published by Elsevier Inc.)

Computer-aided diagnosis systems for assisting the classification of various diseases have the potential to improve radiologists' diagnostic accuracy and efficiency, as reported in several studies. Conventional systems generally provide the probabilities of disease types in terms of numerical values, a method that may not be efficient for radiologists who are trained by reading a large number of images. Presentation of reference images similar to those of a new case being diagnosed can supplement the probability outputs based on computerized analysis as an intuitive guide, and it can assist radiologists in their diagnosis, reporting, and treatment planning. Many studies on content-based medical image retrievals have been reported on. For retrieval of perceptually similar and diagnostically relevant images, incorporation of perceptual similarity data by radiologists has been suggested. In this paper, studies on image retrieval methods are reviewed with a special focus on quantification, utilization, and the evaluation of subjective similarities between pairs of images.

Rationale: Multilineage-differentiating stress enduring (Muse) cells, pluripotent marker stage-specific embryonic antigen-3 + cells, are nontumorigenic endogenous pluripotent-like stem cells obtainable from various tissues including the bone marrow. Their therapeutic efficiency has not been validated in acute myocardial infarction., Objective: The main objective of this study is to clarify the efficiency of intravenously infused rabbit autograft, allograft, and xenograft (human) bone marrow-Muse cells in a rabbit acute myocardial infarction model and their mechanisms of tissue repair., Methods and Results: In vivo dynamics of Nano-lantern-labeled Muse cells showed preferential homing of the cells to the postinfarct heart at 3 days and 2 weeks, with ≈14.5% of injected GFP (green fluorescent protein)-Muse cells estimated to be engrafted into the heart at 3 days. The migration and homing of the Muse cells was confirmed pharmacologically (S1PR2 [sphingosine monophosphate receptor 2]-specific antagonist JTE-013 coinjection) and genetically (S1PR2-siRNA [small interfering ribonucleic acid]-introduced Muse cells) to be mediated through the S1P (sphingosine monophosphate)-S1PR2 axis. They spontaneously differentiated into cells positive for cardiac markers, such as cardiac troponin-I, sarcomeric α-actinin, and connexin-43, and vascular markers. GCaMP3 (GFP-based Ca calmodulin probe)-labeled Muse cells that engrafted into the ischemic region exhibited increased GCaMP3 fluorescence during systole and decreased fluorescence during diastole. Infarct size was reduced by ≈52%, and the ejection fraction was increased by ≈38% compared with vehicle injection at 2 months, ≈2.5 and ≈2.1 times higher, respectively, than that induced by mesenchymal stem cells. These effects were partially attenuated by the administration of GATA4 -gene-silenced Muse cells. Muse cell allografts and xenografts efficiently engrafted and recovered functions, and allografts remained in the tissue and sustained functional recovery for up to 6 months without immunosuppression., Conclusions: Muse cells may provide reparative effects and robust functional recovery and may, thus, provide a novel strategy for the treatment of acute myocardial infarction., (© 2018 American Heart Association, Inc.)

We propose an approach to supporting pre-surgical planning for the uterus by integrating medical image analysis and physical model generation based on 3D printing. With our method, we first segment the patient-specific anatomy and lesions of the uterus on MR images; then, we create a 3D physical model, an exact replica of the patient's uterus in terms of size and softness, with transparency for easy observation of the internal structures of the uterus. In our experiments, we created pre-surgical models of hysterectomy for five patients who had been diagnosed to have uterine endometrial cancer. An experienced radiologist, the surgeons, and all of the patients cooperated in our experiment for carrying out subjective evaluations of the usefulness of our model. The accuracy of the physical models was evaluated quantitatively by comparison between the MR images of the patients and the CT images of the models. The results showed that the mean values of the errors in gap ranged from 1.19 to 2.22 mm, which was satisfactory for the surgeons. The feedback from both surgeons and patients demonstrated the usefulness and convenience of the models for efficient patient explanation understanding and pre-surgical planning by surgeons.

The pattern of blood vessels in the eye is unique to each person because it rarely changes over time. Therefore, it is well known that retinal blood vessels are useful for biometrics. This paper describes a biometrics method using the Jaccard similarity coefficient (JSC) based on blood vessel regions in retinal image pairs. The retinal image pairs were rough matched by the center of their optic discs. Moreover, the image pairs were aligned using the Iterative Closest Point algorithm based on detailed blood vessel skeletons. For registration, perspective transform was applied to the retinal images. Finally, the pairs were classified as either correct or incorrect using the JSC of the blood vessel region in the image pairs. The proposed method was applied to temporal retinal images, which were obtained in 2009 (695 images) and 2013 (87 images). The 87 images acquired in 2013 were all from persons already examined in 2009. The accuracy of the proposed method reached 100%.

Dental records play an important role in forensic identification. To this end, postmortem dental findings and teeth conditions are recorded in a dental chart and compared with those of antemortem records. However, most dentists are inexperienced at recording the dental chart for corpses, and it is a physically and mentally laborious task, especially in large scale disasters. Our goal is to automate the dental filing process by using dental x-ray images. In this study, we investigated the application of a deep convolutional neural network (DCNN) for classifying tooth types on dental cone-beam computed tomography (CT) images. Regions of interest (ROIs) including single teeth were extracted from CT slices. Fifty two CT volumes were randomly divided into 42 training and 10 test cases, and the ROIs obtained from the training cases were used for training the DCNN. For examining the sampling effect, random sampling was performed 3 times, and training and testing were repeated. We used the AlexNet network architecture provided in the Caffe framework, which consists of 5 convolution layers, 3 pooling layers, and 2 full connection layers. For reducing the overtraining effect, we augmented the data by image rotation and intensity transformation. The test ROIs were classified into 7 tooth types by the trained network. The average classification accuracy using the augmented training data by image rotation and intensity transformation was 88.8%. Compared with the result without data augmentation, data augmentation resulted in an approximately 5% improvement in classification accuracy. This indicates that the further improvement can be expected by expanding the CT dataset. Unlike the conventional methods, the proposed method is advantageous in obtaining high classification accuracy without the need for precise tooth segmentation. The proposed tooth classification method can be useful in automatic filing of dental charts for forensic identification., (Copyright © 2016 Elsevier Ltd. All rights reserved.)

Purpose: Studies reported that the mandibular cortical width (MCW) measured on dental panoramic radiographs (DPRs) was significantly correlated with bone mineral density. However, MCW is not a perfect index by itself, and studies suggest the added utility of mandibular cortical index (MCI). In this study, we propose a method for computerized estimation of mandibular cortical degree (MCD), a new continuous measure of MCI, for osteoporotic risk assessment., Methods: The mandibular contour was automatically segmented using an active contour model. The regions of interest near mental foramen were extracted for MCW and MCD determination. The MCW was measured on the basis of residue-line detection results and pixel profiles. Image features including texture features based on gray-level co-occurrence matrices were determined. The MCD were estimated using support vector regression (SVR). The SVR was trained using previously collected 99 DPRs, including 26 osteoporotic cases, by a computed radiography system. The proposed scheme was tested using 99 DPRs obtained by a photon-counting system with data of bone mineral density at distal forearm. The number of osteoporotic, osteopenic, and control cases were 12, 18, and 69 cases, respectively. The subjective MCD by a dental radiologist was employed for training and evaluation., Results: The correlation coefficients with the subjective MCD were -0.549 for MCW alone, 0.609 for the MCD by the features without MCW, and 0.617 for the MCD by the features and MCW. The correlation coefficients with the BMD were 0.619, -0.608, and -0.670, respectively. The areas under the receiver operating characteristic curves for detecting osteoporotic cases were 0.830, 0.884, and 0.901, respectively, whereas those for detecting high-risk cases were 0.835, 0.833, and 0.880, respectively., Conclusions: In conclusion, our scheme may have a potential to identify asymptomatic osteoporotic and osteopenic patients through dental examinations.

Retinal arteriolar narrowing is decided based on the artery and vein diameter ratio (AVR). Previous methods segmented blood vessels and classified arteries and veins by color pixels in the centerlines of blood vessels. AVR was definitively determined through measurement of artery and vein diameters. However, this approach was not sufficient for cases with close contact between the artery of interest and an imposing vein. Here, an algorithm for AVR measurement via new classification of arteries and veins is proposed. In this algorithm, additional steps for an accurate segmentation of arteries and veins, which were not identified using the previous method, have been added to better identify major veins in the red channel of a color image. To identify major arteries, a decision tree with three features was used. As a result, all major veins and 90.9% of major arteries were correctly identified, and the absolute mean error in AVRs was 0.12. The proposed method will require further testing with a greater number of images of arteriolar narrowing before clinical application.

Textural features can be useful in differentiating between benign and malignant breast lesions on mammograms. Unlike previous computerized schemes, which relied largely on shape and margin features based on manual contours of masses, textural features can be determined from regions of interest (ROIs) without precise lesion segmentation. In this study, therefore, we investigated an ROI-based feature, namely, radial local ternary patterns (RLTP), which takes into account the direction of edge patterns with respect to the center of masses for classification of ROIs for benign and malignant masses. Using an artificial neural network (ANN), support vector machine (SVM) and random forest (RF) classifiers, the classification abilities of RLTP were compared with those of the regular local ternary patterns (LTP), rotation invariant uniform (RIU2) LTP, texture features based on the gray level co-occurrence matrix (GLCM), and wavelet features. The performance was evaluated with 376 ROIs including 181 malignant and 195 benign masses. The highest areas under the receiver operating characteristic curves among three classifiers were 0.90, 0.77, 0.78, 0.86, and 0.83 for RLTP, LTP, RIU2-LTP, GLCM, and wavelet features, respectively. The results indicate the usefulness of the proposed texture features for distinguishing between benign and malignant lesions and the superiority of the radial patterns compared with the conventional rotation invariant patterns., (Copyright © 2016 Elsevier Ltd. All rights reserved.)

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.

Detection of lacunar infarcts is important because their presence indicates an increased risk of severe cerebral infarction and dementia. However, accurate identification of lacunar infarcts is often difficult for radiologists. Our previous computer-aided detection (CAD) scheme achieved a sensitivity of 96.8% with 0.76 false positives (FPs) per slice. However, further reduction of FPs remained an issue for the clinical application. The purpose of this study is to improve our CAD scheme by using kernel eigenspace template matching. First, we selected the regions of interest (ROIs) around the candidate regions detected in our previous method. A kernel eigenspace was then made by using kernel principal component analysis of the training data set. A test ROI was projected onto the same kernel eigenspace as the training data set. The cross-correlation coefficients between the test ROI and all the training ROIs were calculated on the kernel eigenspace. By comparing the two maxima of coefficients with a lacunar ROI and an FP ROI, the test ROI was classified. By using the proposed method, the quantity of the templates became 1.9% of that in template matching on the real space and 31. 9% of FPs could be eliminated while keeping the same sensitivity; nevertheless 30.3% of FPs were eliminated when we employed the eigenspace template matching under the same condition. Therefore, kernel eigenspace template matching could improve FP rate without a significant reduction in the true positive rate.

The purpose of this study was to improve an automated scheme for detecting carotid artery calcification (CAC) in dental panoramic radiographs (DPRs). Using 100 DPRs, the sensitivity of CAC detection employing our previous method was 90.0% with 5.0 false positives (FPs) per image. This study describes two enhancements. One is the adoption of a new feature for the position of CACs in addition to previous features. The other is feature selection employing the support vector machine using all combinations. Five of 12 features were selected. Using our proposed method, the average sensitivity for the same database proved to be 90.0%, with only 2.5 FPs per image. These results indicate the potential effectiveness of the new positional feature and feature selection.

Glaucoma is a leading cause of permanent blindness. Retinal imaging is useful for early detection of glaucoma. In order to evaluate the presence of glaucoma, ophthalmologists may determine the cup and disc areas and diagnose glaucoma using a vertical optic cup-to-disc (C/D) ratio and a rim-to-disc (R/D) ratio. Previously we proposed a method to determine cup edge by analyzing a vertical profile of pixel values, but this method provided a cup edge smaller than that of an ophthalmologist. This paper describes an improved method using the locations of the blood vessel bends. The blood vessels were detected by a concentration feature determined from the density gradient. The blood vessel bends were detected by tracking the blood vessels from the disc edge to the primary cup edge, which was determined by our previous method. Lastly, the vertical C/D ratio and the R/D ratio were calculated. Using forty-four images, including 32 glaucoma images, the AUCs of both the vertical C/D ratio and R/D ratio by this proposed method were 0.966 and 0.936, respectively.

Presentation of similar reference images can be useful for diagnosis of new lesions. A similarity map which can visually present the overview of the relationship between the lesions with different types may provide the supplemental information to the reference images. A new method for constructing the similarity map by multidimensional scaling (MDS) for breast masses on mammograms was investigated. Nine pathologic types were included; three regions of interests each from the nine groups were employed in this study. Subjective similarity ratings by expert readers were obtained for all possible 351 pairs of masses. Using the average ratings, MDS similarity map was created. Each axis of the MDS configuration was fitted by the linear model with 13 image features to reconstruct the similarity map. Dissimilarity based on the distance in the reconstructed space was determined and compared with the subjective rating. The MDS map consistently represented the similarity between cysts and fibroadenomas, invasive lobular carcinomas and scirrhous carcinomas, and ductal carcinomas in situ, solid-tubular carcinomas, and papillotubular carcinomas with the experts' data. The correlation between the average subjective ratings and the dissimilarities based on the distance in the reconstructed feature space was much greater (-0.87) than that of the dissimilarities based on the distance in the conventional feature space (-0.65). The new similarity map by MDS can be useful for visualizing the relationship between breast masses with different pathologic types. It has potential usefulness in selecting the similarity measures and providing the supplemental information.

Computer-aided diagnosis has potential in improving radiologists' diagnosis, and presentation of similar images as a reference may provide additional useful information for distinction between benign and malignant lesions. In this study, we evaluated the usefulness of presentation of reference images in observer performance studies and compared the results obtained by groups of observers practicing in the United States and Japan. The results showed that the presentation of the reference images was generally effective for both groups, as the areas under the receiver operating characteristic curves improved from 0.915 to 0.924 for the group in the US and from 0.913 to 0.925 for the group in Japan, although the differences were marginally (p = 0.047) and not (p = 0.13) statistically significant, respectively. There was a slight difference between the two groups in the way that the observers reacted to some benign cases, which might be due to differences in the population of screenees and in the socio-clinical environment. In the future, it may be worthwhile to investigate the development of a customized system for physicians in different socio-clinical environments.

Diabetic retinopathy (DR) is the most frequent cause of blindness. Microaneurysm (MA) is an early symptom of DR. Therefore, the detection of MA is important for the early detection of DR. We have proposed an automated MA detection method based on double-ring filter, but it has given many false positives. In this paper, we propose an MA detection method based on eigenvalue analysis using a Hessian matrix, with an aim to improve MA detection. After image preprocessing, the MA candidate regions were detected by eigenvalue analysis using the Hessian matrix in green-channeled retinal fundus images. Then, 126 features were calculated for each candidate region. By a threshold operation based on feature analysis, false positive candidates were removed. The candidate regions were then classified either as MA or false positive using artificial neural networks (ANN) based on principal component analysis (PCA). The 126 features were reduced to 25 components by PCA, and were then inputted to ANN. When the method was evaluated on visible MAs using 25 retinal images from the retinopathy online challenge (ROC) database, the true positive rate was 73%, with eight false positives per image.

Our motivation was to provide an automatic tool for radiologists and orthopedic surgeons for improving the quality of life of an aging population. We propose a method for generating a shape model and a fully automated segmenting scheme for the psoas major muscle in X-ray CT images by using the shape model. Our approach consists of two steps: (1) The generation of a shape model and its application to muscle segmentation. The shape model describes the muscle's outer shape and has two parameters, an outer shape parameter and a fitting parameter. The former was determined by approximating of the outer shape of the muscle region in training cases. The latter was determined for each test case in the recognition process. (2) Finally, the psoas major muscle was segmented by use of the shape model. To evaluate the performance of the method, we applied it to CT images for constructing the shape models by using 20 cases as training samples; 80 cases were used for testing. The accuracy of this method was measured by comparison of the extracted muscle regions with regions that were identified manually by an expert radiologist. The experimental results of the segmentation of the psoas major muscle gave a mean Jaccard similarity coefficient of 72.3%. The mean true segmentation coefficient was 76.2%. The proposed method can be used for the analysis of cross-sectional area and muscular thickness in a transverse section, offering radiologists an alternative to manual measurement for saving their time and improving the reproducibility of segmentation.

Glaucoma is the first leading cause of vision loss in Japan, thus developing a scheme for helping glaucoma diagnosis is important. For this problem, automated nerve fiber layer defects (NFLDs) detection method was proposed, but glaucoma risk assessment using this method was not evaluated. In this paper, computerized risk assessment for having glaucoma was attempted by use of the patients' clinical information, and the performances of the NFLDs detection and the glaucoma risk assessment were compared. The clinical data includes the systemic data, ophthalmologic data, and right and left retinal images. Glaucoma risk assessment was built by using machine learning technique, which were artificial neural network, radial basis function (RBF) network, k-nearest neighbor algorithm, and support vector machine. The inputting parameter was ten clinical ones with/without the results of NFLDs detection. As a result, proposed glaucoma risk assessment showed the higher performance than the NFLD detection. The result of the glaucoma risk assessment indicates that the computerized assessment may be useful for the determination of glaucoma risk.

An automated method for measurement of arteriolar-to-venular diameter ratio (AVR) is presented. The method includes optic disc segmentation for the determination of the AVR measurement zone, retinal vessel segmentation, vessel classification into arteries and veins, selection of major vessel pairs, and measurement of AVRs. The sensitivity for the major vessels in the measurement zone was 87%, while 93% of them were classified correctly into arteries or veins. In 36 out of 40 vessel pairs, at least parts of the paired vessels were correctly identified. Although the average error in the AVRs with respect to those based on the manual vessel segmentation results was 0.11, the average error in vessel diameter was less than 1 pixel. The proposed method may be useful for objective evaluation of AVRs and has a potential for detecting focal arteriolar narrowing on macula-centered screening fundus images., (Copyright © 2011 Elsevier Ltd. All rights reserved.)

Retinal image examination is useful for early detection of glaucoma, which is a leading cause of permanent blindness. In order to evaluate the presence of glaucoma, ophthalmologists may determine the cup and disc areas and diagnose glaucoma using a vertical cup-to-disc ratio. However, determination of the cup area based on computation algorithm is very difficult, thus we propose a method to measure the cup-to-disc ratio using a vertical profile on the optic disc. The edge of optic disc was then detected by use of a Canny edge detection filter. The profile was then obtained around the center of the optic disc. Subsequently, the edges of the cup area were determined by classification of the profiles based on zero-crossing method. Lastly, the vertical cup-to-disc ratio was calculated. Using forty five images, including twenty three glaucoma images, the AUC of 0.947 was achieved with this method.

Presence of peripapillary chorioretinal atrophy (PPA) is considered one of the risk factors for glaucoma. It can be identified as bright regions in retinal fundus images, and therefore, incorrectly included as the part of the optic disc regions in the automated disc detection scheme. For potential risk assessment and use in improving optic disc segmentation, a computerized detection of PPA was investigated. By using texture analysis, the sensitivity for detecting the moderate to severe PPA regions in the test dataset was 73% with the specificity of 95%. The proposed method may be useful for identifying the cases with the PPA in retinal fundus images.

The automatic determination of the optic disc area in retinal fundus images can be useful for calculation of the cup-to-disc (CD) ratio in the glaucoma screening. We compared three different methods that employed active contour model (ACM), fuzzy c-mean (FCM) clustering, and artificial neural network (ANN) for the segmentation of the optic disc regions. The results of these methods were evaluated using new databases that included the images captured by different camera systems. The average measures of overlap between the disc regions determined by an ophthalmologist and by using the ACM (0.88 and 0.87 for two test datasets) and ANN (0.88 and 0.89) methods were slightly higher than that by using FCM (0.86 and 0.86) method. These results on the unknown datasets were comparable with those of the resubstitution test; this indicates the generalizability of these methods. The differences in the vertical diameters, which are often used for CD ratio calculation, determined by the proposed methods and based on the ophthalmologist's outlines were even smaller than those in the case of the measure of overlap. The proposed methods can be useful for automatic determination of CD ratios., (2011 Elsevier Ireland Ltd. All rights reserved.)

The purpose of this study is to develop an automated carotid artery calcification (CAC) detection scheme on dental panoramic radiographs (DPRs). The CAC is one of the indices for predicting the risk of arteriosclerosis. First, regions of interest (ROIs) that include CACs were determined on the basis of inflection points of the mandibular contour. Initial CAC candidates were detected by using a grayscale top-hat filter and simple grayscale thresholding technique. Finally, a rule-based approach and support vector machine to reduce the number of false positive (FP) findings were applied using features such as area, location, and circularity. Thirty-four DPRs were used to evaluate the proposed scheme. The sensitivity for the detection of CACs was 93.6% with 4.4 FPs per image. Experimental results showed that our computer-aided detection scheme may be useful to detect CACs.

The effect of the presentation of similar images for distinction between benign and malignant masses on mammograms was evaluated in the observer performance study. Images of masses were obtained from the Digital Database for Screening Mammography. We selected 50 benign and 50 malignant masses by a stratified randomization method. For each case, similar images were selected based on the size of masses and the similarity measures. Radiologists were shown images with unknown masses and asked to provide their confidence level that the lesions were malignant before and after the presentation of the similar images. Eleven observers, including three attending breast radiologists, three breast imaging fellows, and five residents, participated. The average areas under the receiver operating characteristic curves without and with the presentation of the similar images were almost equivalent. However, there were many cases in which the similar images caused beneficial effects to the observers, whereas there were a small number of cases in which the similar images had detrimental effects. From a detailed analysis of the reasons for these detrimental effects, we found that the similar images would not be useful for diagnosis of rare and very difficult cases, i.e., benign-looking malignant and malignant-looking benign cases. In addition, these cases should not be included in the reference database, because radiologists would be confused by these unusual cases. The results of this study could be very important and useful for the future development and improvement of a computer-aided diagnosis system.

Rationale and Objectives: To determine similarity measures for selection of pathology-known similar images that would be useful for radiologists as a reference guide in the diagnosis of new breast lesions on mammograms., Materials and Methods: The images were obtained from the Digital Database for Screening Mammography developed by the University of South Florida. For determination and evaluation of similarity measures, the "gold standard" of similarities for 300 pairs of masses was determined by 10 breast radiologists. For determining similarity measures that would agree with radiologists' similarity determination, an artificial neural network (ANN) was trained with the radiologists' subjective similarity ratings and the image features. The image features were determined subjectively using the Breast Imaging Reporting and Data System (BI-RADS) lesion descriptors and objectively by computerized image analysis. The similarity measures determined by the ANN were compared to the gold standard and evaluated in terms of the correlation coefficient., Results: The similarity measures determined using the BI-RADS descriptors only were not as useful as those determined by use of the image features only. When the BI-RADS margin ratings were combined with the image features, the correlation coefficient between the subjective ratings and the objective measures improved slightly (r = 0.76) compared to those based on the image features alone (r = 0.74)., Conclusions: The inclusion of the BI-RADS margin descriptors may be useful for determination of similarity measures, especially when it is difficult to obtain the manual outlines of the masses and if the BI-RADS descriptors were provided consistently by radiologists.

Rationale and Objectives: We conducted an observer study to investigate whether radiologists can judge similarities in pairs of breast masses and lung nodules consistently and reproducibly., Materials and Methods: Institutional review board approval and informed observer consent were obtained. This study was compliant with the Health Insurance Portability and Accountability Act. We used eight pairs of breast masses on mammograms and eight pairs of lung nodules on computed tomographic images, for which subjective similarity ratings ranging from 0 to 1 were determined in our previous studies. From these, four sets of image pairs were created (ie, a set of eight mass pairs, a set of eight nodule pairs, and two mixed sets of four mass and four nodule pairs). Eight radiologists, including four breast radiologists and four chest radiologists, compared all combinations of the eight pairs in each set using a two-alternative forced-choice (2AFC) method to determine the similarity ranking scores by identifying which pair was more similar than the other pair based on the overall impression for diagnosis., Results: In the mass set and nodule set, the relationship between the average subjective similarity ratings and the average similarity ranking scores by 2AFC indicated very high correlations (r = 0.91 and 0.88). Moreover, in the two mixed sets, the correlations between the average subjective similarity ratings and the average similarity ranking scores were also very high (r = 0.90 and 0.98). Thus, radiologists were able to compare the similarities for pairs of lesions consistently, even in the unusual comparison of pairs of completely different types of lesions., Conclusion: The subjective similarity of a pair of lesions in medical images can be quantified consistently by a group of radiologists. The concept of similarity of lesions in medical images can be subjected to rigorous scientific research and investigation in the future.

This work presents the usefulness of texture features in the classification of breast lesions in 5,518 images of regions of interest, which were obtained from the Digital Database for Screening Mammography that included microcalcifications, masses, and normal cases. Sixteen texture features were used, i.e., 13 were based on the spatial gray-level dependence matrix and 3 on the wavelet transform. The nonparametric K-NN classifier was used in the classification stage. The results obtained from receiver operating characteristic analysis indicated that the texture features can be used for separating normal regions and lesions with masses and microcalcifications, yielding the area under the curve (AUC) values of 0.957 and 0.859, respectively. However, the texture features were not very effective for distinguishing between malignant and benign lesions because the AUC was 0.617 for masses and 0.607 for microcalcifications. The study showed that the texture features can be used for the detection of suspicious regions in mammograms.

We investigated a psychophysical similarity measure for selection of images similar to those of unknown masses on mammograms, which may assist radiologists in the distinction between benign and malignant masses. Sixty pairs of masses were selected from 1445 mass images prepared for this study, which were obtained from the Digital Database for Screening Mammography by the University of South Florida. Five radiologists provided subjective similarity ratings for these 60 pairs of masses based on the overall impression for diagnosis. Radiologists' subjective ratings were marked on a continuous rating scale and quantified between 0 and 1, which correspond to pairs not similar at all and pairs almost identical, respectively. By use of the subjective ratings as "gold standard," similarity measures based on the Euclidean distance between pairs in feature space and the psychophysical measure were determined. For determination of the psychophysical similarity measure, an artificial neural network (ANN) was employed to learn the relationship between radiologists' average subjective similarity ratings and computer-extracted image features. To evaluate the usefulness of the similarity measures, the agreement with the radiologists' subjective similarity ratings was assessed in terms of correlation coefficients between the average subjective ratings and the similarity measures. A commonly used similarity measure based on the Euclidean distance was moderately correlated (r=0.644) with the radiologists' average subjective ratings, whereas the psychophysical measure by use of the ANN was highly correlated (r=0.798). The preliminary result indicates that a psychophysical similarity measure would be useful in the selection of images similar to those of unknown masses on mammograms., (© 2005 American Association of Physicists in Medicine.)