1,721,122 research outputs found
A deep convolutional neural network for breast density assessment: Optimization and explainability.
No full textThis work aims to develop a method for deep neural network explainability. It is the ability to explain the algorithm behaviour and its predictions when it has a deep multi-layer nonlinear structure. This is a critical issue in Artificial Intelligence. An already developed deep Residual Convolutional Neural Network is able to automatically classify mammograms into breast density classes. The explainability of the network has been studied through various analyses and visualization techniques, assessing trust in the model, which is fundamental for its potential application in clinical practice, and also achieving a performance improvement in terms of accuracy
Explainability of a CNN for breast density assessment
No full textDeep neural network explainability is a critical issue in Artificial Intelligence (AI). This work aims to develop a method to explain a deep residual Convolutional Neural Network able to automatically classify mammograms into breast density classes. Breast density, a risk factor for breast cancer, is defined as the amount of fibroglandular tissue compared to fat tissue visible on a mammogram. We studied the explainability of the classifier to understand the reasons behind its predictions, in fact with a deep multi-layer structure, it acts like a black-box. As there is no well-established method, we explored different possible analyses and visualization techniques. The main obtained results were the achievement of a performance improvement in terms of accuracy and a contribution to assess trust in the model. This is fundamental for a potential application in clinical practice
An automatic system to discriminate malignant from benign massive lesions in mammograms
No full textEvaluating the degree of malignancy of a massive lesion on the basis of the mere visual analysis of the mammogram is a non-trivial task. We developed a semi-automated system for massive-lesion characterization with the aim to support the radiological diagnosis. A dataset of 226 masses has been used in the present analysis. The system performances have been evaluated in terms of the area under the ROC curve, obtaining Az= 0.80 ± 0.04
An automatic system to discriminate malignant from benign massive lesions in mammograms
No full textEvaluating the degree of malignancy of a massive lesion on the basis of the mere visual analysis of the mammogram is a non-trivial task. We developed a semi-automated system for massive-lesion characterization with the aim to support the radiological diagnosis. A dataset of 226 masses has been used in the present analysis. The system performances have been evaluated in terms of the area under the ROC curve, obtaining Az= 0.80 ± 0.04
MRIndex: A tool for evaluating muscle involvement in neuromuscular diseases from MRI images
No full textThe progress and severity of neuromuscular conditions can be monitored in several ways, most of which are invasive and, thus, poorly acceptable, for the patient. Among the least obtrusive solutions, the use of muscular magnetic resonance imaging is gaining importance in last decades, therefore becoming the elective methodology for defining the muscular involvement in such conditions. However, subjectivity is always quite frequent in the interpretation of biomedical images, and the diagnosis is often demanded to the experience of the clinician's eye. With MRIndex, a novel tool for the automated analysis of muscular magnetic resonance images, a quantitative 'picture' of the muscular involvement in neuromuscular conditions is defined, stemming from the definition of fat infiltration within the muscular tissue, well-known biomarker for disease severity. The solution is currently used at clinics and preliminary at-a-glance results are quite promising. If such evidence is confirmed on larger cohorts and with a more robust statistical approach, the tool can represent a groundbreaking alternative in current clinical practice, possibly bringing to a more objective definition of the clinical status of a patient and helping in the personalization of the treatment, possibly improving their outcome
Convolutional neural networks for breast density classification: Performance and explanation insights
No full textWe propose and evaluate a procedure for the explainability of a breast density deep learning based classifier. A total of 1662 mammography exams labeled according to the BI-RADS categories of breast density was used. We built a residual Convolutional Neural Network, trained it and studied the responses of the model to input changes, such as different distributions of class labels in training and test sets and suitable image pre-processing. The aim was to identify the steps of the analysis with a relevant impact on the classifier performance and on the model explainability. We used the grad-CAM algorithm for CNN to produce saliency maps and computed the Spearman’s rank correlation between input images and saliency maps as a measure of explanation accuracy. We found that pre-processing is critical not only for accuracy, precision and recall of a model but also to have a reasonable explanation of the model itself. Our CNN reaches good performances compared to the state-of-art and it considers the dense pattern to make the classification. Saliency maps strongly correlate with the dense pattern. This work is a starting point towards the implementation of a standard framework to evaluate both CNN performances and the explainability of their predictions in medical image classification problems
Residual convolutional neural networks to automatically extract significant breast density features
No full textIn this paper, we present a work on breast density classification performed with deep residual neural network and we discuss the future analysis we could perform. Breast density is one of the most important breast cancer risk factor and it represents the amount of fibroglandular tissue with respect to fat tissue as seen on a mammographic exam. However, it is not easy to include it in risk models because of its variability among women and its qualitative definition. We trained a deep CNN to perform breast density classification in two ways. First, we classified mammograms using two “super-classes” that are dense and non-dense breast. Second, we trained the residual neural network to classify mammograms according to the four classes of the BI-RADS standard. We obtained very good results compared to our literature knowledge in terms of accuracy and recall. In the near future, we are going to improve the robustness of our algorithm with respect to the mammographic systems used and we want to include pathological exams too. Then we want to study and characterize the CNN-extracted features in order to identify the most significant for breast density. Finally, we want to study how to quantitatively measure the precision of the network in capturing the significative part of the images
Comprehensive assessment of image quality in synthetic and digital mammography: a quantitative comparison
No full textRecent advances in digital breast tomosynthesis (DBT) technology were focused on the reconstruction of 2D “Synthesized Mammograms” (SMs) from DBT dataset. The introduction of SMs could avoid an additional digital mammography (DM) which is often required in complement to DBT examinations. Therefore, breast absorbed dose and compression time can be significantly reduced in DBT+SM procedures with respect to DBT+DM modality. However, to date, a limited number of studies have objectively characterised the image quality of SMs with respect to DM images. Therefore, the aim of this phantom study was to comprehensively compare SMs and DM images in terms of several image quality parameters. A Selenia Dimensions system (Hologic, Bedford, Mass, USA) was employed in this work. Five different phantoms were adopted to study noise, contrast and spatial resolution properties of the images. Specifically, noise power spectrum (NPS), maps of signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), modulation transfer function (MTF) and contrast-detail (CD) thresholds were evaluated both for SM and DM modalities. SMs were characterised by different texture, noise and SNR spatial distribution properties with respect to DM images. Additionally, while in some conditions SM provides higher CNR than DM, lower overall performances in terms spatial resolution and CD curves were found in comparison to DM images. Therefore, given the great benefits of SMs in terms of dose and compression time saving, further clinical investigations on SMs image quality properties could be of practical interest to integrate our findings
Comparison of radiation dose between standard 2D Full-Field Digital Mammography (FFDM) and 3D Digital Breast Tomosynthesis (DBT) by using a dose monitoring system.
No full textThe aim of the study is to compare the radiation dose of a clinical system for the acquisition of mammographic and tomosynthesis images. About 4000 exams are performed every year on adult female patients using the same device (Selenia Dimensions; Hologic, Bedford, MA, USA) that provides digital 2D and 3D images of the breast. We present the dose assessment performed by means of a radiation dose-tracking tool developed in our department that extracts data from the DICOM Header archived in our PACS. Average Glandular Dose (AGD) evaluated by the system is extracted by our dose-tracking tool
Average Absorbed Breast Dose (2ABD) to Mean Glandular Dose (MGD) Conversion Function for Digital Breast Tomosynthesis: A New Approach
No full textBackground: In this work a new method for the Mean Glandular Dose evaluation in digital breast tomosynthesis (DBT) is presented. Methods: Starting from the experimental-based dosimetric index, 2ABD, which represents the average absorbed breast dose, the mean glandular dose MGD2ABD was calculated using a conversion function of glandularity f(G), obtained through the use of Monte Carlo simulations. Results: f(G) was computed for a 4.5 cm thick breast: from its value MGD2ABD for different compressed breast thicknesses and glandularities was obtained. The comparison between MGD2ABD estimates and the dosimetric index provided in the current dosimetry protocols, following the Dance's approach, MGDDance, showed a good agreement (<10%) for all the analyzed breast thicknesses and glandularities. Conclusion: The strength of the proposed method can be considered an accurate mean glandular dose assessment starting from few and accessible parameters, reported in the header DICOM of each DBT exam
- …
