Background Artificial intelligence (AI) is increasingly used to manage radiologists' workloads. The impact of patient characteristics on AI performance has not been well studied. Purpose To understand the (race and ethnicity, age, breast density) an algorithm interpreting negative screening digital tomosynthesis (DBT) examinations. Materials Methods This retrospective cohort study identified DBT examinations from academic institution January 1, 2016, December 31, 2019. All had 2 years...

Detecting an anomaly such as a malignant tumor or nodule from medical images including mammogram, CT PET is still ongoing research problem drawing lot of attention with applications in diagnosis. A conventional way to address this learn discriminative model using training datasets negative and positive samples. The learned can be used classify testing sample into class. However, applications, the high unbalance between samples poses difficulty for learning algorithms, they will biased...

Purpose To develop a deep learning algorithm that uses temporal information to improve the performance of previously published framework cancer lesion detection for digital breast tomosynthesis. Materials and Methods This retrospective study analyzed current 1-year-prior Hologic tomosynthesis screening examinations from eight different institutions between 2016 2020. The dataset contained 973 7123 noncancer cases. front end this was an existing performed single-view followed by ipsilateral...

In mammography, calcifications are one of the most common signs breast cancer. Detection such lesions is an active area research for computer-aided diagnosis and machine learning algorithms. Due to limited numbers positive cases, many supervised detection models suffer from overfitting fail generalize. We present a one-class, semi-supervised framework using deep convolutional autoencoder trained with over 50,000 images 11,000 negative-only cases. Since model learned only normal parenchymal...

Interpretability in machine learning models is important high-stakes decisions, such as whether to order a biopsy based on mammographic exam. Mammography poses challenges that are not present other computer vision tasks: datasets small, confounding information present, and it can be difficult even for radiologist decide between watchful waiting mammogram alone. In this work, we framework interpretable learning-based mammography. addition predicting lesion malignant or benign, our work aims...

Computer-aided detection (CAD) frameworks for breast cancer screening have been researched several decades. Early adoption of deep-learning models in CAD has shown greatly improved performance compared to traditional on single-view images. Recently, studies by merging information from multiple views within each exam. Clinically, the integration lesion correspondence during is a complicated decision process that depends correct execution referencing steps. However, most multi-view are...

Purpose: To determine whether domain transfer learning can improve the performance of deep features extracted from digital mammograms using a pre-trained convolutional neural network (CNN) in prediction occult invasive disease for patients with ductal carcinoma situ (DCIS) on core needle biopsy. Method: In this study, we collected mammography magnification views 140 DCIS at biopsy, 35 which were subsequently upstaged to cancer. We utilized CNN model that was two natural image data sets...

Our goal is to develop a 2.5D CNN model detect multiple diseases in organs CT scans. In this study we investigated detection of 4 common the lungs, which are atelectasis, edema, pneumonia and nodule. Most existing algorithms for computer-aided diagnosis (CAD) use 2D models axial slices. hypothesis that by using information from all three views (coronal, sagittal axial), may achieve better classification result, because some be more obvious different view or combination multi-views. data...

Recent advancements in conditional Generative Adversarial Networks (cGANs) have shown promises label guided image synthesis. Semantic masks, such as sketches and maps, are another intuitive effective form of guidance Directly incorporating the semantic masks constraints dramatically reduces variability quality synthesized results. We observe this is caused by incompatibility features from different inputs (such mask latent vector) generator. To use whilst providing realistic results with...

In previous work, we generated computational breast phantoms by using a principal component analysis (PCA) or "Eigenbreast" technique. For this study, sought to address resolution limitations in the synthesized analyzing new human subject data set with higher resolution. We utilize PCA sample input cases, then weighted sums along different eigenvectors "eigenbreasts," number of cases can be generated. While breasts vary structure and form, used series compressed derived from CT volumes...

Digital breast tomosynthesis (DBT), synthetic mammography, and full-field digital mammography (FFDM) are commonly used medical imaging modalities for cancer screening. Due to the data complexity, most CAD research applies only one modality, which under-utilizes complementary information in these 2D 3D modalities. In this study, we propose a Residual-Attention Multimodal Fusion network (ResAMF-Net) that integrates lesion features across We evaluated performance on large private dataset,...

Detection of suspicious breast cancer lesion in screening mammography images is an important step for the downstream diagnosis cancer. A trained radiologist can usually take advantage multi-view correlation lesions to locate abnormalities. In this work, we investigate feasibility using a random image pair same from exam detection lesions. We present novel approach utilize single shot system inspired by You only look once (YOLO) v1 simultaneously process primary view and secondary...

When we deploy machine learning models in high-stakes medical settings, must ensure these make accurate predictions that are consistent with known science. Inherently interpretable networks address this need by explaining the rationale behind each decision while maintaining equal or higher accuracy compared to black-box models. In work, present a novel neural network algorithm uses case-based reasoning for mammography. Designed aid radiologist their decisions, our presents both prediction of...

Detection and localization of microcalcification (MC) clusters are very important in mammography diagnosis. Supervised MC detectors require learning from extracted individual MCs clusters. However, they limited by number datasets given that images hard to obtain. In this work, we propose a method detect malignant using unsupervised, one-class, deep convolutional autoencoder. Specifically, designed autoencoder model where only patches normal cases' mammograms used during training. We then...

Annotated data availability has always been a major limiting f actor for the development of algorithms in field computer aided diagnosis. The purpose this study is to investigate feasibility using conditional generative adversarial network (GAN) synthesize high resolution mammography images with semantic control. We feed binary mammographic texture map generator full-field digital-mammogram (FFDM). Our results show quickly learned grow anatomical details around edges within mask. However, we...

Detecting microcalcification clusters in mammograms is important to the diagnosis of breast diseases. Previous studies which mainly focused on supervised methods require abundant annotated training data but these are usually hard acquire. In this work, we leverage unsupervised convolutional autoencoders and structural similarity (SSIM) based post-processing detect localize full-field digital (FFDMs). Our models were trained by patches extracted from 3,632 normal cases, total with 16,702...

Most of the existing CAD frameworks for digital breast tomosynthesis (DBT) are single-view only, while radiologists typically utilize information from multiple screening views to better detect cancer lesions. Previously, we developed Retina-Match framework lesion detection that performed ipsilateral matching between CC and MLO same breast. In this work, improve in both sampling strategy feature extraction processes. We proposed a "hard negative" train on more difficult pairs increase...