Computer-aided assessment of breast density: comparison of supervised deep learning and feature-based statistical learning

Breast density is one of the most significant factors for cancer risk. In this study, we proposed a supervised deep learning approach for automated estimation of percentage density (PD) on digital mammography (DM). The deep convolutional neural network (DCNN) was trained to estimate a probability map of breast density (PMD). PD was calculated as the ratio of the dense area to the breast area based on the probability of each pixel belonging to dense region or fatty region at a decision threshold of 0.5. The DCNN estimate was compared to a feature-based statistical learning approach, in which gray level, texture and morphological features were extracted from each ROI and the least absolute shrinkage and selection operator (LASSO) was used to select and combine the useful features to generate the PMD. The reference PD of each image was provided by two experienced MQSA radiologists. With IRB approval, we retrospectively collected 347 DMs from patient files at our institution. The 10-fold cross-validation results showed a strong correlation r=0.96 between the DCNN estimation and interactive segmentation by radiologists while that of the feature-based statistical learning approach vs radiologists’ segmentation had a correlation r=0.78. The difference between the segmentation by DCNN and by radiologists was significantly smaller than that between the feature-based learning approach and radiologists (p < 0.0001) by two-tailed paired t-test. This study demonstrated that the DCNN approach has the potential to replace radiologists’ interactive thresholding in PD estimation on DMs.

Recommending music based on a user’s music preference is a way to improve user listening experience. Finding the correlation between the user data (e.g., location, time of the day, music listening history, emotion, etc.) and the music is a challenging task. In this paper, we propose an emotion-aware personalized music recommendation system (EPMRS) to extract the correlation between the user data and the music. To achieve this correlation, we combine the outputs of two approaches: the deep convolutional neural networks (DCNN) approach and the weighted feature extraction (WFE) approach. The DCNN approach is used to extract the latent features from music data (e.g., audio signals and corresponding metadata) for classification. In the WFE approach, we generate the implicit user rating for music to extract the correlation between the user data and the music data. In the WFE approach, we use the term-frequency and inverse document frequency (TF-IDF) approach to generate the implicit user ratings for the music. Later, the EPMRS recommends songs to the user based on calculated implicit user rating for the music. We use the million songs dataset (MSD) to train the EPMRS. For performance comparison, we take the content similarity music recommendation system (CSMRS) as well as the personalized music recommendation system based on electroencephalography feedback (PMRSE) as the baseline systems. Experimental results show that the EPMRS produces better accuracy of music recommendations than the CSMRS and the PMRSE. Moreover, we build the Android and iOS APPs to get realistic data of user experience on the EPMRS. The collected feedback from anonymous users also show that the EPMRS sufficiently reflect their preference on music.

This paper presents a model of Deep Convolutional Neural Networks (CNN) based on transfer learning for image recognition. This means to use a Deep CNN system pretrained on the large ImageNet dataset of 14 million images and 1000 classes in order to learn feature selection. The results of the pretraining phase are transferred to the problem of classification for the images belonging to the UC Merced Land Use dataset with 21 classes. As benchmark, we have considered a Deep CNN trained with a fraction of the same UC Merced Land Use dataset containing the test images for classification. The experimental results have pointed out the obvious advantage of the Deep CNN with transfer learning (accuracy of 0.87 using pretraining over 0.46 for fully training on the same dataset).

BackgroundOsteoporosis, a disease stemming from bone metabolism irregularities, affects approximately 200 million people worldwide. Timely detection of osteoporosis is pivotal in grappling with this public health challenge. Deep learning (DL), emerging as a promising methodology in the field of medical imaging, holds considerable potential for the assessment of bone mineral density (BMD). This study aimed to propose an automated DL framework for BMD assessment that integrates localization, segmentation, and ternary classification using various dominant convolutional neural networks (CNNs).MethodsIn this retrospective study, a cohort of 2,274 patients underwent chest computed tomography (CT) was enrolled from January 2022 to June 2023 for the development of the integrated DL system. The study unfolded in 2 phases. Initially, 1,025 patients were selected based on specific criteria to develop an automated segmentation model, utilizing 2 VB-Net networks. Subsequently, a distinct cohort of 902 patients was employed for the development and testing of classification models for BMD assessment. Then, 3 distinct DL network architectures, specifically DenseNet, ResNet-18, and ResNet-50, were applied to formulate the 3-classification BMD assessment model. The performance of both phases was evaluated using an independent test set consisting of 347 individuals. Segmentation performance was evaluated using the Dice similarity coefficient; classification performance was appraised using the receiver operating characteristic (ROC) curve. Furthermore, metrics such as the area under the curve (AUC), accuracy, and precision were meticulously calculated.ResultsIn the first stage, the automatic segmentation model demonstrated excellent segmentation performance, with mean Dice surpassing 0.93 in the independent test set. In the second stage, both the DenseNet and ResNet-18 demonstrated excellent diagnostic performance in detecting bone status. For osteoporosis, and osteopenia, the AUCs were as follows: DenseNet achieved 0.94 [95% confidence interval (CI): 0.91–0.97], and 0.91 (95% CI: 0.87–0.94), respectively; ResNet-18 attained 0.96 (95% CI: 0.92–0.98), and 0.91 (95% CI: 0.87–0.94), respectively. However, the ResNet-50 model exhibited suboptimal diagnostic performance for osteopenia, with an AUC value of only 0.76 (95% CI: 0.69–0.80). Alterations in tube voltage had a more pronounced impact on the performance of the DenseNet. In the independent test set with tube voltage at 100 kVp images, the accuracy and precision of DenseNet decreased on average by approximately 14.29% and 18.82%, respectively, whereas the accuracy and precision of ResNet-18 decreased by about 8.33% and 7.14%, respectively.ConclusionsThe state-of-the-art DL framework model offers an effective and efficient approach for opportunistic osteoporosis screening using chest CT, without incurring additional costs or radiation exposure.

We are developing a decision support system for treatment response assessment of bladder cancer in CT urography (CTU). Accurate segmentation of bladder cancer is a critical and challenging task. We previously developed a bladder cancer segmentation method using a deep learning convolutional neural network and level set (DL-CNN+LS) approach. In this study we investigated the application of a U-Net based deep learning (U-Net) model for bladder cancer segmentation. Our new U-Net method did not require the second-stage level set refinement, greatly simplifying the overall segmentation pipeline. The proposed U-Net model utilized a user-defined box to direct the attention of the U-Net to the lesion region by masking out the structured background outside the box. We trained and evaluated the performance of the U-Net segmentations by using hand-drawn 3D contours from a radiologist as reference standard. The segmentation accuracy was evaluated by the average volume intersection ratio (AVI), average percent volume error (AVE), average absolute volume error (AAVE), average minimum distance (AMD), and the Jaccard index (JI). On the validation set, the cropped U-Net achieved values of AVI = 65.5&plusmn;20.2%, AVE = 3.7&plusmn;44.4%, AAVE = 30.9±30.8%, AMD = 2.9&plusmn;1.5 mm, and JI = 51.4&plusmn;16.8%. The previous DL-CNN+LS approach achieved values of AVI = 36.2&plusmn;28.4%, AVE = 51.1&plusmn;39.3%, AAVE = 52.0&plusmn;38.0%, AMD = 4.1&plusmn;1.7 mm, and JI = 32.9&plusmn;27.2%. On the independent test set, the cropped U-Net achieved values of AVI = 66.0&plusmn;22.8%, AVE = -3.7&plusmn;39.4%, AAVE = 28.1&plusmn;27.3%, AMD = 4.5&plusmn;3.3 mm, and JI = 49.2&plusmn;20.0%. The DL-CNN+LS achieved values of AVI = 31.2&plusmn;24.3%, AVE = 59.9&plusmn;30.9%, AAVE = 60.8&plusmn;29.0%, AMD = 5.5&plusmn;2.2 mm, and JI = 27.9&plusmn;20.8%, respectively. The results demonstrated that the U-Net model could achieve a higher accuracy than the previous DL-CNN+LS model while reducing the complexity of the segmentation pipeline.

Deep learning has emerged as a state-of-the-art learning technique across a wide range of applications, including image recognition, object detection and localisation, natural language processing, prediction and forecasting systems. With significant applicability, deep learning could be used in new and broader areas of applications, including remanufacturing. Remanufacturing is a process of taking used products through disassembly, inspection, cleaning, reconditioning, reassembly and testing to ascertain that their condition meets new products conditions with warranty. This process is complex and requires a good understanding of the respective stages for proper analysis. Inspection is a critical process in remanufacturing, which guarantees the quality of the remanufactured products. It is currently an expensive manual operation in the remanufacturing process that depends on operator expertise, in most cases. This research investigates the application of deep learning algorithms to inspection in remanufacturing, towards automating the inspection process. This paper presents a novel vision-based inspection system based on deep convolution neural network (DCNN) for eight types of defects, namely pitting, rust, cracks and other combination faults. The materials used for this feasibility study were 100 cm × 150 cm mild steel plate material, purchased locally, and captured using a USB webcam of 0.3 megapixels. The performance of this preliminary study indicates that the DCNN can classify with up to 100% accuracy on validation data and above 96% accuracy on a live video feed, by using 80% of the sample dataset for training and the remaining 20% for testing. Therefore, in the remanufacturing parts inspection, the DCNN approach has high potential as a method that could surpass the current technologies used in the design of inspection systems. This research is the first to apply deep learning techniques in remanufacturing inspection. The proposed method offers the potential to eliminate expert judgement in inspection, save cost, increase throughput and improve precision. This preliminary study demonstrates that deep learning techniques have the potential to revolutionise inspection in remanufacturing. This research offers valuable insight into these opportunities, serving as a starting point for future applications of deep learning algorithms to remanufacturing.

Diabetic Retinopathy (DR) is eye disease caused through the retinal damage induced through long-term diabetes mellitus illness. In the initial stage, DR has minor vision difficulties or no symptoms but it outputs in vision loss if it is not treated early. However, overfitting occurs when a model learns noise or specific features from a small dataset which leading to poor generalization. To solve this issue, this research proposed an Otsu Multi-level Adaptive Thresholding with Deep Convolution Neural Network (OMAT+DCNN) approach for DR segmentation and classification. The Contrast Limited Adaptive Histogram Equalization (CLAHE) is used for preprocessing input image which enhances the contrast level and image visibility without amplifying noise. The OMAT is used for segmenting the affected regions and ResNet50 is used for extracting Deep Learning (DL) features. Then, relevant features are selected through Multi-Verse Optimizer (MVO) and DCNN is used for classification. The OMAT improves segmentation accuracy by considering multiple thresholds which adopts to local intensity variations within the image, even when the dataset is limited. DCNN capturing intricate patterns and spatial hierarchies within images and its deep architecture allows to learn complex representations which leads to enhance accuracy. The parameters like dice coefficient, Intersection over Union (IoU), Mean IoU (MIoU), precision, accuracy, f1-score and sensitivity are used for assessing performance. The OMAT+DCNN reaches 0.993 of accuracy when compared to existing techniques such as Convolutional Transformer Network (CTNet) and Residual Attention Network (RAN).

The feasibility of localizing, segmenting, and classifying individual cells in multi-channel confocal microscopy images is highly dependent on image quality. In certain applications, with good image quality, the segmentation can be trivial and accomplished via thresholding, watershed, or a collection of other well-established and studied heuristics; however, at the limit of poor image quality and complex image features, these techniques fail. It is at this limit that deep convolutional neural network (DCNN) approaches excel. Our research studies the interaction of individual immune cells and their shape changes relative to inflammatory immune reactions1 using multi-channel immunofluorescence imaging of renal biopsies from patients with inflammatory kidney disease. We present here a deep learning methodology for application to nuclear and cell membrane immunofluorescent stains to automatically segment and classify multiple T-cell and dendritic cell types. With both T-cells and dendritic cells segmented, we are able to study how T-cells with different surface antigens change shape with proximity to dendritic cells. Shape changes are seen when T-cells move close to dendritic cells and interact. We use a sliding window, max-filtering DCNN to segment and classify 3 cell types from 6 image stains channels within a single DCNN. This DCNN maintains images at original resolution throughout the network using dilated convolutions and max-filtering in place of max pooling layers. In addition, we use 3D convolution kernels with two spatial dimensions and one channel dimension. This allows us to output a multi-class binary classification based on multichannel data at the original image resolution. We trained and validated the network across 24 patients with 8,572 segmented cells. Our results demonstrate a mean Dice-Sorensen score of 0.78 +/- 0.18, a mean classification sensitivity of 0.76, and a mean classification specificity of 0.75 across all 3 segmented cell types.

Deep learning can be used to classify images to verify or correct DICOM header information. One situation where this is useful is in the classification of thoracic radiographs that were acquired anteroposteriorly (AP) or posteroanteriorly (PA). A convolutional neural network (CNN) was previously trained and showed a strong performance in the task of classifying between AP and PA radiographs, giving a 0.97 ± 0.005 AUC for an independent test set. However, 81% of the AP training set and 24% of the AP independent test set consisted of images with imprinted labels. To evaluate the effect of labels on training and testing of a CNN, the labels on the images used for training were removed by cropping. Then the CNN was retrained using the cropped images with the same training parameters as before. The retrained CNN was tested on the same independent test set and resulted in a 0.95 ± 0.007 AUC in the task of classifying between AP and PA radiographs. The p-value is 0.002 between the AUCs from the two networks, showing a statistically significant decrease in performance by the network trained on the cropped images. The decrease in performance may be due to the network being previously trained to recognize imprinted labels or due to relevant anatomy being cropped along with the label, however, the performance is still high and can be incorporated in clinical workflow.

Background/Objectives: 123I-FP-CIT dopamine transporter imaging is commonly used for the diagnosis of Parkinsonian syndromes in patients whose clinical presentation is atypical. Prolonged immobility, which can be difficult to maintain in this population, is required to perform SPECT acquisition. In this study we aimed to develop a Convolutional Neural Network (CNN) able to predict the outcome of the full examination based on the first acquired projection, and reliably detect normal patients. Methods: All 123I-FP-CIT SPECT performed in our center between June 2017 and February 2024 were included and split between a training and a validation set (70%/30%). An additional 100 SPECT were used as an independent test set. Examinations were labeled by two independent physicians. A VGG16-like CNN model was trained to assess the probability of examination abnormality from the first acquired projection (anterior and posterior view at 0°), taking age into consideration. A threshold maximizing sensitivity while maintaining good diagnostic accuracy was then determined. The model was validated in the independent testing set. Saliency maps were generated to visualize the most impactful areas in the classification. Results: A total of 982 123I-FP-CIT SPECT were retrieved and labelled (training set: 618; validation set: 264; independent testing set: 100). The trained model achieved a sensibility of 98.0% and a negative predictive value of 96.3% (one false negative) while maintaining an accuracy of 75.0%. The saliency maps confirmed that the regions with the greatest impact on the final classification corresponded to clinically relevant areas (basal ganglia and background noise). Conclusions: Our results suggest that this trained CNN could be used to exclude presynaptic dopaminergic loss with high reliability from the first acquired projection. It could be particularly useful in patients with compliance issues. Confirmation with images from other centers will be necessary.

Although CT radiomics has shown promising results in the evaluation of vertebral fractures, the need for manual segmentation of fractured vertebrae limited the routine clinical implementation of radiomics. Therefore, automated segmentation of fractured vertebrae is needed for successful clinical use of radiomics. In this study, we aimed to develop and validate an automated algorithm for segmentation of fractured vertebral bodies on CT, and to evaluate the applicability of the algorithm in a radiomics prediction model to differentiate benign and malignant fractures. A convolutional neural network was trained to perform automated segmentation of fractured vertebral bodies using 341 vertebrae with benign or malignant fractures from 158 patients, and was validated on independent test sets (internal test, 86 vertebrae [59 patients]; external test, 102 vertebrae [59 patients]). Then, a radiomics model predicting fracture malignancy on CT was constructed, and the prediction performance was compared between automated and human expert segmentations. The algorithm achieved good agreement with human expert segmentation at testing (Dice similarity coefficient, 0.93–0.94; cross-sectional area error, 2.66–2.97%; average surface distance, 0.40–0.54 mm). The radiomics model demonstrated good performance in the training set (AUC, 0.93). In the test sets, automated and human expert segmentations showed comparable prediction performances (AUC, internal test, 0.80 vs 0.87, p = 0.044; external test, 0.83 vs 0.80, p = 0.37). In summary, we developed and validated an automated segmentation algorithm that showed comparable performance to human expert segmentation in a CT radiomics model to predict fracture malignancy, which may enable more practical clinical utilization of radiomics.

Histopathological grading of breast cancer is an important tumor-related prognostic factor and plays an important role in breast cancer prognosis analysis. Nowadays, histopathological grading of breast cancer is mainly identified by pathological images and radiologists cannot differentiate the histopathological grade directly from digital mammograms. In this paper, we propose to discriminate the histopathological grades directly based on digital mammograms, which is noninvasive and convenient. End-to-end training Convolutional Neural Network (CNN) is firstly designed to extract semantic features directly from raw image data. Considering the scarce annotated mammograms data and large size of tumor region, a light and deep network with less training parameters is modified to prevent overfitting. Results demonstrate that our proposed network is superior to other CNN models and traditional classifier based on hand-crafted features.

Convolutional neural networks (CNNs) show potential for delineating cancers on contrast-enhanced MRI (ce-MRI) but there are clinical scenarios in which administration of contrast is not desirable. We investigated performance of the CNN for delineating primary nasopharyngeal carcinoma (NPC) on non-contrast-enhanced images and compared the performance to that on ce-MRI. We retrospectively analyzed primary NPC in 195 patients using a well-established CNN, U-Net, for tumor delineation on the non-contrast-enhanced fat-suppressed (fs)-T2W, ce-T1W and ce-fs-T1W images. The CNN-derived delineations were compared to manual delineations to obtain Dice similarity coefficient (DSC) and average surface distance (ASD). The DSC and ASD on fs-T2W were compared to those on ce-MRI. Primary tumor volumes (PTVs) of CNN-derived delineations were compared to that of manual delineations. The CNN for NPC delineation on fs-T2W images showed similar DSC (0.71 ± 0.09) and ASD (0.21 ± 0.48cm) to those on ce-T1W images (0.71 ± 0.09 and 0.17 ± 0.19cm, respectively) (p > 0.05), and lower DSC but similar ASD to ce-fs-T1W images (0.73 ± 0.09, p < 0.001; and 0.17 ± 0.20cm, p > 0.05). The CNN overestimated PTVs on all sequences (p < 0.001). The CNN showed promise for NPC delineation on fs-T2W images in cases where it is desirable to avoid contrast agent injection. The CNN overestimated PTVs on all sequences.

Breast density and background parenchymal enhancement (BPE) are suggested to be related to the risk of breast cancer. The first step to quantitative analysis of breast density and BPE is segmenting the breast from body. Nowadays, convolutional neural networks (CNNs) are widely used in image segmentation and work well in semantic segmentation, however, CNNs have been rarely used in breast region segmentation. In this paper, the CNN was employed to segment the breast region in transverse fat-suppressed breast dynamic contrast enhanced magnetic resonance imaging (DCE-MRI). Image normalization was initially performed. Subsequently, the dataset was divided into three sets randomly: train set validation set and test set. The 2-D U-Net was trained by train set and the optimum model was chosen by validation set. Finally, segmentation results of test set obtained by U-Net were adjusted in the postprocessing. In this step, two largest volumes were computed to determine whether the smaller volume is the scar after mastectomy. With the limitation of small dataset, 5-fold cross-validation and data augmentation were used in this study. Final results on the test set were evaluated by volume-based and boundary-based metrics with manual segmentation results. By using this method, the mean dice similarity coefficient (DSC), dice difference coefficient (DDC), and root-mean-square distance reached 97.44%, 5.11%, and 1.25 pixels, respectively.

In modern manufacturing the ability of retracing produced components is crucial for quality management and process optimization. Tracking is essential, especially for analyzing the influence of the production parameters on the final quality of the castings. In the iron foundry industry, common marking methods, such as a datamatrix code, cannot be used due to harsh environmental conditions and the rough surface of the cast parts. This work presents a new coding and reading system that guarantees unique marking in the casting process.The coding is built up over several beveled pins and is read out using an optical 2D handheld scanner. With a deep convolutional neural network approach of object detection and classification, a stable image processing algorithm is presented. With a first prototype a reading accuracy of 99.86% for each pin was achieved with an average scanning time of 0.43 s. The presented code is compatible with existing foundry processes, while the handheld scanner is intuitive and reliable. This allows immediate benefits for process optimization.