Ella Barkan

A Region Based Convolutional Network for Tumor Detection and Classification in Breast Mammography

This paper addresses the problem of detection and classification of tumors in breast mammograms. We introduce a novel system that integrates several modules including a breast segmentation module and a fibroglandular tissue segmentation module into a modified cascaded region-based convolutional network. The method is evaluated on a large multi-center clinical dataset and compared to ground truth annotated by expert radiologists. Preliminary experimental results show the high accuracy and efficiency obtained by the suggested network structure. As the volume and complexity of data in healthcare continues to accelerate generalizing such an approach may have a profound impact on patient care in many applications.

A Generic Form Processing Approach for Large Variant Templates

In today’s world, form processing systems must be able to recognize mutant forms that appear to be based on differing templates but are actually only a variation of the original. A single definition of a representative template actually covers large varieties of the same logical templates. We developed a method and system, similar to the human visual system, which differentiates between templates via features such as logos, dominant words, and geometrical shapes, while ignoring minor details and variations. When the system finds an appropriate template, it then decodes the content of the form. Our approach has been applied in several scenarios with encouraging results.

Medical Image Description Using Multi-task-loss CNN

Automatic detection and classification of lesions in medical images remains one of the most important and challenging problems. In this paper, we present a new multi-task convolutional neural network (CNN) approach for detection and semantic description of lesions in diagnostic images. The proposed CNN-based architecture is trained to generate and rank rectangular regions of interests (ROI’s) surrounding suspicious areas. The highest score candidates are fed into the subsequent network layers. These layers are trained to generate semantic description of the remaining ROI’s.

Detection and Segmentation of Antialiased Text in Screen Images

Various software applications deal with analyzing the textual content of screen captures. Interpreting these images as text poses several challenges, relative to images traditionally handled by optical character recognition (OCR) engines. One such challenge is caused by text antialiasing, a technique which blurs the edges of characters, to reduce jagged appearance. This blurring changes the character images according to context, and can sometimes fuse them together. In this paper, we offer a low-cost method that can be used as a preprocessing stage, prior to OCR. Our method locates antialiased text in a screen image and segments it into separate character images. Our proposed algorithm significantly improves OCR results, particularly in images with colored text of small font size, such as in graphic user interface (GUI) screens.

The scanline principle: efficient conversion of display algorithms into scanline mode

The scanline principle is a general technique for efficiently converting any display algorithm that is based on polygon scan conversion into scanline mode, i.e., the image is produced in scanline order with required memory proportional to one scanline. Based on critical-points scan conversion, the technique reduces the Z-buffer or its variants to one scanline. Current scanline depth buffers are inefficient in both time and space. The scanline principle can also transform listpriority methods, such as BSP trees, into scanline mode. The scanline mode enables efficient supersampling and averaging, and low latency in image generation, compression and transmission.

Mammogram Classification and Abnormality Detection from Nonlocal Labels using Deep Multiple Instance Neural Network

Mammography is the common modality used for screening and early detection of breast cancer. The emergence of machine learning, particularly deep learning methods, aims to assist radiologists to reach higher sensitivity and specificity. Yet, typical supervised machine learning methods demand the radiological images to have findings annotated within the image. This is a tedious task, which is often out of reach due to the high cost and unavailability of expert radiologists. We describe a computeraided detection and diagnosis system for weakly supervised learning, where the mammogram (MG) images are tagged only on a global level, without local annotations. Our work addresses the problem of MG classification and detection of abnormal findings through a novel deep learning framework built on the multiple instance learning (MIL) paradigm. Our proposed method processes the MG image utilizing the full resolution, with a deep MIL convolutional neural network. This approach allows us to classify the whole MG according to a severity score and localize the source of abnormality in full resolution, while trained on a weakly labeled data set. The key hallmark of our approach is automatic discovery of the discriminating patches in the mammograms using MIL. We validate the proposed method on two mammogram data sets, a large multi-center MG cohort and the publicly available INbreast, in two different scenarios. We present promising results in classification and detection, comparable to a recent supervised method that was trained on fully annotated data set. As the volume and complexity of data in healthcare continues to increase, such an approach may have a profound impact on patient care in many applications.

A CNN based method for automatic mass detection and classification in mammograms

Abstract A novel system for detection and classification of masses in breast mammography is introduced. The system integrates a breast segmentation module together with a modified region-based convolutional network to obtain detection and classification of masses according to BI-RADS score. While most of the previous work on mass identification in breast mammography has focused on classification, this study proposes to solve both the detection and the classification problems. The method is evaluated on a large multi-centre clinical data-set and compared to ground truth annotated by expert radiologists. Preliminary experimental results show the high accuracy and efficiency obtained by the suggested network structure. As the volume and complexity of data in health care continues to accelerate generalising such an approach may have a profound impact on patient care in many applications.

Siamese Network for Dual-View Mammography Mass Matching

In a standard mammography screening procedure, two X-ray images are acquired per breast from two views. In this paper, we introduce a patch based, deep learning network for lesion matching in dual-view mammography using a Siamese network. Our method is evaluated on several datasets, among them the large freely available digital database for screening mammography (DDSM). We perform a comprehensive set of experiment, focusing on the mass correspondence problem. We analyze the effect of transfer learning between different types of dataset, compare the network based matching to classic template matching and evaluate the contribution of the matching network to the detection task. Experimental results show the promise in improving detection accuracy by our approach.

Deep Learning for Automatic Detection of Abnormal Findings in Breast Mammography

Automatic identification of abnormalities is a key problem in medical imaging. While the majority of previous work in mammography has focused on classification of abnormalities rather than detection and localization, here we introduce a novel deep learning method for detection of masses and calcifications. The power of this approach comes from generating an ensemble of individual Faster-RCNN models each trained for a specific set of abnormal clinical categories, together with extending a modified two stage Faster-RCNN scheme to a three stage cascade. The third stage being an additional classifier working directly on the image pixels with the handful of sub-windows generated by the first two stages. The performance of the algorithm is evaluated on the INBreast benchmark and on a large internal multi-center dataset. Quantitative results compete well with state of the art in terms of accuracy. Computationally the methods runs significantly faster than current state-of-the art techniques.

CP3: Robust, Output‐sensitive Display of Convex Polyhedra in Scanline Mode

A new technique is developed for displaying disjoint convex polyhedra. The method has the following properties: It is output‐sensitive, displays the objects in scanline mode, and it is naturally robust. There is no complex data structure uniting the different polyhedra, so dynamic insertions and deletions are simple. Its robustnes is based on a novel method of comparing depths by representative “axes” of objects instead of surfaces. The method is based on two extensions of the “critical‐points” method for polygon scan conversion: One extension allows the efficient display of planar graphs in scanline mode, and another extension is into the third dimension. Test runs indicate that it compares extremely favorably with other methods that operate in scanline mode, as well as with standard software and hardware techniques of medium‐level workstations.