Digital Mammogram Images Research Articles

Breast cancer characterization using region-based convolutional neural network with screening and diagnostic mammogram

Background: Detection and classification of microcalcifications in breast tissues is crucial for early breast cancer diagnosis and long-term treatment. Objective: This paper aims to propose a robust model capable of detection and classification of breast cancer calcifications in digital mammogram images using Deep Convolutional Neural Networks (DCNN). Materials and methods: An expert breast radiologist annotated the 3,265 clinical mammogram images to create a comprehensive ground truth dataset comprising 2,500 annotations for malignant and benign calcifications. This dataset was utilized to train our model, a two-stage detection system incorporating a Region-based Convolutional Neural Network (RCNN) with AlexNet and support vector machines to enhance the system’s robustness. The proposed model was compared to the one-stage detection, utilizing YOLOv4 combined with the Cross-Stage Partial Darknet53 (CSPDarknet53) architecture. A separate dataset of 504 mammogram images was explicitly set aside for model testing. The efficacy of the proposed model was evaluated based on key performance metrics, including precision, recall, F1 score, and mean average precision (mAP). Results: The results showed that the proposed RCNN-2 model could automatically identify and categorize calcifications as malignant or benign, outperforming the YOLOv4 models. The RCNN-2’s overall effectiveness, as evaluated by precision, recall, F1 score, and mean average precision (mAP), achieved scores of 0.82, 0.85, 0.83, and 0.74, respectively. Conclusion: The proposed RCNN-2 model demonstrates very effective detection and classification of calcification in mammogram images, especially in high-dense breast images. The performance of the proposed model was compared to that of YOLOv4, and it can be concluded that the proposed RCNN model yields outstanding performance. The model can be a helpful tool for radiologists.

Multi-class Breast Cancer Classification Using CNN Features Hybridization

Breast cancer has become the leading cause of cancer mortality among women worldwide. The timely diagnosis of such cancer is always in demand among researchers. This research pours light on improving the design of computer-aided detection (CAD) for earlier breast cancer classification. Meanwhile, the design of CAD tools using deep learning is becoming popular and robust in biomedical classification systems. However, deep learning gives inadequate performance when used for multilabel classification problems, especially if the dataset has an uneven distribution of output targets. And this problem is prevalent in publicly available breast cancer datasets. To overcome this, the paper integrates the learning and discrimination ability of multiple convolution neural networks such as VGG16, VGG19, ResNet50, and DenseNet121 architectures for breast cancer classification. Accordingly, the approach of fusion of hybrid deep features (FHDF) is proposed to capture more potential information and attain improved classification performance. This way, the research utilizes digital mammogram images for earlier breast tumor detection. The proposed approach is evaluated on three public breast cancer datasets: mammographic image analysis society (MIAS), curated breast imaging subset of digital database for screening mammography (CBIS-DDSM), and INbreast databases. The attained results are then compared with base convolutional neural networks (CNN) architectures and the late fusion approach. For MIAS, CBIS-DDSM, and INbreast datasets, the proposed FHDF approach provides maximum performance of 98.706%, 97.734%, and 98.834% of accuracy in classifying three classes of breast cancer severities.

In situ tumor model for longitudinal in silico imaging trials

Objective. In this article, we introduce a computational model for simulating the growth of breast cancer lesions accounting for the stiffness of surrounding anatomical structures. Approach. In our model, ligaments are classified as the most rigid structures while the softer parts of the breast are occupied by fat and glandular tissues As a result of these variations in tissue elasticity, the rapidly proliferating tumor cells are met with differential resistance. It is found that these cells are likely to circumvent stiffer terrains such as ligaments, instead electing to proliferate preferentially within the more yielding confines of the breast’s soft topography. By manipulating the interstitial tumor pressure in direct proportion to the elastic constants of the tissues surrounding the tumor, this model thus creates the potential for realizing a database of unique lesion morphology sculpted by the distinctive topography of each local anatomical infrastructure. We modeled the growth of simulated lesions within volumes extracted from fatty breast models, developed by Graff et al with a resolution of 50 μm generated with the open-source and readily available Virtual Imaging Clinical Trials for Regulatory Evaluation (VICTRE) imaging pipeline. To visualize and validate the realism of the lesion models, we leveraged the imaging component of the VICTRE pipeline, which replicates the siemens mammomat inspiration mammography system in a digital format. This system was instrumental in generating digital mammogram (DM) images for each breast model containing the simulated lesions. Results. By utilizing the DM images, we were able to effectively illustrate the imaging characteristics of the lesions as they integrated with the anatomical backgrounds. Our research also involved a reader study that compared 25 simulated DM regions of interest (ROIs) with inserted lesions from our models with DM ROIs from the DDSM dataset containing real manifestations of breast cancer. In general the simulation time for the lesions was approximately 2.5 hours, but it varied depending on the lesion’s local environment. Significance. The lesion growth model will facilitate and enhance longitudinal in silico trials investigating the progression of breast cancer.

Texture features analysis technique to detect mass lesion in digitized mammogram images

Mass lesions are one of the breast cancer tumors. Mammogram images are the first screening tool to detect tumors in the women breast, but due to radiologist fatigue, number of false positive (FP) and false negative (FN) rates are increased. The main objective of this paper is to develop an intelligent computer aided diagnosis (CAD) system that can accurately detect mass lesions in digitized mammogram images. The proposed method has three stages. The first stage is a preprocessing stage, where the mass lesion is enhanced using a customized Laplacian filter. Then, multi-statistical filters are implemented to detect a potential mass lesion in the mammogram images. In the final stage, the number detected FP regions are reduced using five texture features. The proposed algorithm is evaluated using 45 mammogram images and the algorithm achieved an accuracy rate of 97% in detecting mass lesion with 83% sensitivity rate and 98% specificity rate.

Modelling of Hybrid Meta heuristic Based Parameter Optimizers with Deep Convolutional Neural Network for Mammogram Cancer Detection

Breast cancer (BC) is the common type of cancer among females. Mortality from BC could be decreased by identifying and diagnosing it atan earlierphase. Different imaging modalities are used to detect BC, like mammography. Even withproven records as a BC screening tool, mammography istime-consuming and hasconstraints, namely lower sensitivity in women with dense breast tissue. Computer-Aided Diagnosis or Detection (CAD) system assistsaproficient radiologist to identifyBC at an earlier stage. Recently, the advancementin deep learning (DL)methodsareemployed to mammography assist radiologists to increase accuracy and efficiency. Therefore, this study presents a metaheuristic-based hyperparameter optimization with deep learning-based breast cancer detection on mammogram images (MHODL-BCDMI) technique. The presented MHODL-BCDMI technique mainly focused on the recognition and classification of breast cancer on digital mammograms. To achieve this, the MHODL-BCDMI technique employs pre-processing in two stages: Wiener Filter (WF) based noise elimination and contrast enhancement. Besides, the MHODL-BCDMI technique exploits densely connected networks (DenseNet201) model for feature extraction purposes. For BC classification and detection, a hybrid convolutional neural network with a gated recurrent unit (HCNN-GRU) model is used. Furthermore, three hyperparameter optimizers are employed namely cat swarm optimization (CSO), harmony search algorithm (HSA), and hybrid grey wolf whale optimization algorithm (HGWWOA). Finally, the U2Net segmentation approach is used for the classification of benign and malignant types of cancer. The experimental analysis of the MHODL-BCDMI method is tested on a digital mammogram image dataset and the outcomes are assessed in terms of diverse metrics. The simulation results highlighted the enhanced cancer detection performance of the MHODL-BCDMI technique over other recent algorithms.

Hardware deployment of deep learning model for classification of breast carcinoma from digital mammogram images.

Cancer is an illness that instils fear in many individuals throughout the world due to its lethal nature. However, in most situations, cancer may be cured if detected early and treated properly. Computer-aided diagnosis is gaining traction because it may be used as an initial screening test for many illnesses, including cancer. Deep learning (DL) is a CAD-basedartificial intelligence (AI) powered approach which attempts to mimic the cognitive process of the human brain. Various DL algorithms have been applied for breast cancer diagnosis and haveobtained adequate accuracy due to the DL technology's high feature learning capabilities. However, when it comes to real-time application, deep neural networks (NN) have a high computational complexity in terms of power, speed, and resource usage. With this in mind, this work proposes aminiaturised NN to reduce the number of parameters and computational complexity for hardware deployment. The quantised NN is then accelerated using field-programmable gate arrays (FPGAs) to increase detection speed and minimise power consumption while guaranteeing high accuracy, thusproviding a new avenue in assisting radiologists in breast cancer diagnosis using digital mammograms. When evaluated on benchmark datasets such as DDSM, MIAS, and INbreast, the suggested method achieves high classification rates. The proposed model achieved an accuracy of 99.38% on the combined dataset.

Processing of digital mammogram images using optimized ELM with deep transfer learning for breast cancer diagnosis

Processing of digital mammogram images using optimized ELM with deep transfer learning for breast cancer diagnosis

Deep Learning-Based Metaheuristic Weighted K-Nearest Neighbor Algorithm for the Severity Classification of Breast Cancer

ObjectiveThe most widespread and intrusive cancer type among women is breast cancer. Globally, this type of cancer causes more mortality among women, next to lung cancer. This made the researchers to focus more on developing effective Computer-Aided Detection (CAD) methodologies for the classification of such deadly cancer types. In order to improve the rate of survival and earlier diagnosis, an optimistic research methodology is required in the classification of breast cancer. Consequently, an improved methodology that integrates the principle of deep learning with metaheuristic and classification algorithms is proposed for the severity classification of breast cancer. Hence to enhance the recent findings, an improved CAD methodology is proposed for redressing the healthcare problem. Material and MethodsThe work intends to cast a light-of-research towards classifying the severities present in digital mammogram images. For evaluating the work, the publicly available MIAS, INbreast, and WDBC databases are utilized. The proposed work employs transfer learning for extricating the features. The novelty of the work lies in improving the classification performance of the weighted k-nearest neighbor (wKNN) algorithm using particle swarm optimization (PSO), dragon-fly optimization algorithm (DFOA), and crow-search optimization algorithm (CSOA) as a transformation technique i.e., transforming non-linear input features into minimal linear separable feature vectors. ResultsThe results obtained for the proposed work are compared then with the Gaussian Naïve Bayes and linear Support Vector Machine algorithms, where the highest accuracy for classification is attained for the proposed work (CSOA-wKNN) with 84.35% for MIAS, 83.19% for INbreast, and 97.36% for WDBC datasets respectively. ConclusionThe obtained results reveal that the proposed Computer-Aided-Diagnosis (CAD) tool is robust for the severity classification of breast cancer.

Classification of Multi-view Digital Mammogram Images Using SMO-WkNN

Breast cancer (BCa) is a leading cause of death in the female population across the globe. Approximately 2.3 million new BCa cases are recorded globally in females, overtaking lung cancer as the most prevalent form of cancer to be diagnosed. However, the mortality rates for cervical and BCa are significantly higher in developing nations than in developed countries. Early diagnosis is the only option to minimize the risks of BCa. Deep learning (DL)-based models have performed well in image processing in recent years, particularly convolutional neural network (CNN). Hence, this research proposes a DL-based CNN model to diagnose BCa from digitized mammogram images. The main objective of this research is to develop an accurate and efficient early diagnosis model for BCa detection. This proposed model is a multi-view-based computer-aided diagnosis (CAD) model, which performs the diagnosis of BCa on multi-views of mammogram images like medio-lateral-oblique (MLO) and cranio-caudal (CC). The digital mammogram images are collected from the digital database for screening mammography (DDSM) dataset. In preprocessing, median filter and contrast limited adaptive histogram equalization (CLAHE) techniques are utilized for image enhancement. After preprocessing, the segmentation is performed using the region growing (RG) algorithm. The feature extraction is carried out from the segmented images using a pyramidal histogram of oriented gradients (PHOG) and the AlextNet model. Finally, the classification is performed using the weighted k-nearest neighbor (WkNN) optimized with sequential minimal optimization (SMO). The classified images are evaluated based on accuracy, recall, precision, specificity, f1-score, and mathews correlation coefficient (MCC). Additionally, the false positive and error rates are evaluated. The proposed model obtained 98.57% accuracy, 98.61% recall, 99.25% specificity, 98.63% precision, 97.93% f1-score, 96.26% MCC, 0.0143 error rate, and 0.0075 false positive rate (FPR). Compared to the existing models, the research model has obtained better performances and outperformed the other models.

Evolutionary Computing Model for Finding Breast Cancer Masses using Image Enhancement Procedures with Artificial Intelligent Algorithms

In this research, Particle Swarm Optimization (PSO) based image equalization is projected to enhance the contrast of different breast cancer images. Breast cancer is the highest and another important root of tumor disease in females worldwide. Mass and microcalcification clusters are a significant early signs of breast cancer. The mortality rate can effectively be decreased by early diagnosis and treatment. Most practical approach for the early detection and identification of breast cancer diseases is mammography. Mammographic images contaminated by noise usually involve image enhancement techniques to aid interpretation. Contrast enhancement is divided into two categories: development of direct contrast and enhancement of indirect contrast. Indirect contrast improvement is used in the image histogram update. Histogram Equalization (HE) is the modest enhancement of the indirect contrast approach usually used for contrast enhancement. The proposed method's average entropy is 5.3251 with the highest structural similarity index 0.99725. The best contrast improvement of this method is 1.0404 and PSNR is 46.3803. The MSE value is 2157.08. This paper recommends an innovative method of enhancing digital mammogram image contrast based on different histogram equalization approaches. The performance of the projected method has been related to other prevailing techniques using the parameters, namely, discrete entropy, contrast improvement index, structural similarity index measure, mean square error, and peak signal-to-noise ratio. Investigational findings indicate that the projected strategy is efficient and robust and shows better results than others.

Imaging Modalities in Inflammatory Breast Cancer (IBC) Diagnosis: A Computer-Aided Diagnosis System Using Bilateral Mammography Images.

Inflammatory breast cancer (IBC) is an aggressive type of breast cancer. It leads to a significantly shorter survival than other types of breast cancer in the U.S. The American Joint Committee on Cancer (AJCC) defines the diagnosis based on specific criteria. However, the clinical presentation of IBC in North Africa (Egypt, Morocco, and Tunisia) does not agree, in many cases, with the AJCC criteria. Healthcare providers with expertise in IBC diagnosis are limited because of the rare nature of the disease. This paper reviewed current imaging modalities for IBC diagnosis and proposed a computer-aided diagnosis system using bilateral mammograms for early and improved diagnosis. The National Institute of Cancer in Egypt provided the image dataset consisting of IBC and non-IBC cancer cases. Type 1 and Type 2 fuzzy logic classifiers use the IBC markers that the expert team identified and extracted carefully. As this research is a pioneering work in its field, we focused on breast skin thickening, its percentage, the level of nipple retraction, bilateral breast density asymmetry, and the ratio of the breast density of both breasts in bilateral digital mammogram images. Granulomatous mastitis cases are not included in the dataset. The system's performance is evaluated according to the accuracy, recall, precision, F1 score, and area under the curve. The system achieved accuracy in the range of 92.3-100%.

A Diagnostic Model of Breast Cancer Based on Digital Mammogram Images Using Machine Learning Techniques

Breast cancer disease is one of the most recorded cancers that lead to morbidity and maybe death among women around the world. Recent research statistics have exposed that one from 8 females in the USA and one from 10 females in Europe are contaminated by breast cancer. The challenge with this disease is how to develop a relaxed and fast diagnosing method. One of the attractive ways of early breast cancer diagnosis is based on the mammogram images analysis of the breast using a computer-aided diagnosing (CAD) tool. This paper firstly aimed to propose an efficient method for diagnosing tumors based on mammogram images of breasts using a machine learning approach. Secondly, this paper aimed to the development of a CAD software program for breast cancer diagnosing based on the proposed method in the first step. The followed step-by-step procedure of the proposed method is performed by passing the Mammographic Image Analysis Society (MIAS) through five steps of image preprocessing, image segmentation using seeded region growing (SRG) algorithm, feature extraction using different feature’s extraction classes, and important and effectiveness feature selection using the Sequential Forward Selection (SFS) technique, and finally, the Support Vector Machine (SVM) algorithm is used as a binary classifier in two classification levels. The first level classifier is used to categorize the given image as normal or abnormal while the second-level classifier is used for further classifying the abnormal image as either a malignant or benign cancer. The proposed method is studied and investigated in two phases: the training phase and the testing phase, with the MIAS dataset of mammogram images, using 70% and 30% ratios of dataset images for the training and testing sets, respectively. The practical implementation of the proposed method and the graphical user interface (GUI) CAD tool are carried out using MATLAB software. Experimental results of the proposed method have shown that the accuracy of the proposed method reached 100% in classifying images as normal and abnormal mammogram images while the classification accuracy for benign and malignant is equal to 87.1%.

Deep learning of longitudinal mammogram examinations for breast cancer risk prediction

Information in digital mammogram images has been shown to be associated with the risk of developing breast cancer. Longitudinal breast cancer screening mammogram examinations may carry spatiotemporal information that can enhance breast cancer risk prediction. No deep learning models have been designed to capture such spatiotemporal information over multiple examinations to predict the risk. In this study, we propose a novel deep learning structure, LRP-NET, to capture the spatiotemporal changes of breast tissue over multiple negative/benign screening mammogram examinations to predict near-term breast cancer risk in a case-control setting. Specifically, LRP-NET is designed based on clinical knowledge to capture the imaging changes of bilateral breast tissue over four sequential mammogram examinations. We evaluate our proposed model with two ablation studies and compare it to three models/settings, including 1) a “loose” model without explicitly capturing the spatiotemporal changes over longitudinal examinations, 2) LRP-NET but using a varying number (i.e., 1 and 3) of sequential examinations, and 3) a previous model that uses only a single mammogram examination. On a case-control cohort of 200 patients, each with four examinations, our experiments on a total of 3200 images show that the LRP-NET model outperforms the compared models/settings.

Intelligent computer aided diagnosis system to enhance mass lesions in digitized mammogram images

<span>The paper presents an intelligent system to enhance mass lesions in digitized mammogram images. This system can assist radiologists in detecting mass lesions in mammogram images as an early diagnosis of breast cancer. In this paper, the early detection of mass lesion is visually detected by enhancing mass lesions in mammogram images using hybrid neuro-fuzzy technique. Fuzzified engine is proposed as a first step to convert all pixels in mammogram image to a fuzzy value using three linguistic labels. After that, artificial neural networks are used instead of the inference engine to accurately detect the mass lesions in the mammogram images in a short time. Finally, five linguistic labels are used as a defuzzifier engine to restore the mammogram image. Processed mammogram images are extensively evaluated using two different types of mammogram resources, mammographic image analysis society (MIAS) and University of South Florida (USF) databases. The results show that the proposed intelligent computer aided diagnosis system can successfully enhance the mass lesions in mammogram images with minimum number of false positive regions.</span>

Computerized Analysis of Mammogram Images for Early Detection of Breast Cancer

Breast cancer is widespread worldwide and can be cured if diagnosed early. Using digital mammogram images and image processing with artificial intelligence can play an essential role in breast cancer diagnosis. As many computerized algorithms for breast cancer diagnosis have significant limitations, such as noise handling and varying or low contrast in the images, it can be difficult to segment the abnormal region. These challenges could be overcome by proposing a new pre-processing model, exploring its impact on the post-processing module, and testing it on an extensive database. In this research work, the three-step method is proposed and validated on large databases of mammography images. The first step corresponded to the database classification, followed by the second step, which removed the pectoral muscle from the mammogram image. The third stage utilized new image-enhancement techniques and a new segmentation module to detect abnormal regions in a well-enhanced image to diagnose breast cancer. The pre-and post-processing modules are based on novel image processing techniques. The proposed method was tested using data collected from different hospitals in the Qassim Health Cluster, Qassim Province, Saudi Arabia. This database contained the five categories in the Breast Imaging and Reporting and Data System and consisted of 2892 images; the proposed method is analyzed using the publicly available Mammographic Image Analysis Society database, which contained 322 images. The proposed method gives good contrast enhancement with peak-signal to noise ratio improvement of 3 dB. The proposed method provides an accuracy of approximately 92% on 2892 images of Qassim Health Cluster, Qassim Province, Saudi Arabia. The proposed method gives approximately 97% on the Mammographic Image Analysis Society database. The novelty of the proposed work is that it could work on all Breast Imaging and Reporting and Data System categories. The performance of the proposed method demonstrated its ability to improve the diagnostic performance of the computerized breast cancer detection method.

Breast cancer segmentation using K-means clustering and optimized region-growing technique

Breast cancer is one of the major causes of death among women, and early detection may decrease the aggressiveness of the disease. The goal of this paper is to create an automated system that can classify digital mammogram images into benign and malignant. This paper presents a new detection technique of micro-calcifications in mammogram images. An automated technique for identifying breast microcalcifications (MCs) proposed utilizing two-level segmentation processes, first crop the breast area from the image using k-means clustering, then, an optimized region growing (ORG) approach has been used, where multi-seed points and thresholds are generated optimally depending on the color values of the image pixels. Then the texture features are extracted based on Haralick definitions of texture analysis. In addition, three features (cross-correlation coefficient, pearson correlation, and average area of segmented spots) are obtained from the segmented image. Support vector machine (SVM) classifier evaluate the efficiency of the system utilizing the images from the digital database for screening mammography (DDSM) dataset. The results were obtained by utilizing 355 images for training and 85 images for testing. The proposed system's sensitivity reached up to 97.05%, the specificity obtained is 98.52%, and accuracy is 98.2%.

An Introspective Performance Analysis of Threshold-based Segmentation Techniques on Digital Mammograms

Image segmentation, as a pre-processing step, plays a vital role in medical image analysis. The variants of threshold-based image segmentation methods are proved to offer feasible and optimal solutions to extract the region of interest (RoI), from medical images. Digital mammograms are used as a reliable source of breast cancer prognosis and diagnosis. Thresholding is a simple and effective strategy that finds applications in image processing and analysis. This research aimed to analyze the performance behaviour of a few threshold-based segmentation methods with respect to the intensity distribution of the input mammograms. For this analytical research, six automated thresholding segmentation techniques were chosen: Kapur, Otsu’s, Isoentropic, Ridler & Calvard’s, Kittler & Illingworth's, and Yen. The performance and behaviour of those methods were validated on the digital mammogram images of mini-MIAS database featured with Fatty (F), Fatty-Glandular (G), and Dense-Glandular (D) breast tissues. Those methods were analyzed on two metrics viz., Region Non-Uniformity (RNU) and computation time. The results of this research confirm that Ridler & Calvard’s method gives the best segmentation results for Dense-Glandular, Isoentropic method gives better segmentation results for Fatty and Yen method works well on the Fatty-Glandular normal mammogram images.

Using FC-CLAHE-ADF to enhance digital mammograms for detecting breast lesions

Digital mammograms are commonly used for breast screening, especially to aid the detection of cancer. However, digital mammograms suffer with low contrast images due to the low exposure factors used. This paper presents a novel image enhancement technique for digital mammogram images known as Fuzzy Clipped Contrast-Limited Adaptive Histogram Equalization with Anisotropic Diffusion Filter (FC-CLAHE-ADF). This proposed FC-CLAHE-ADF has adopted a fuzzy-based and histogram-based image enhancement technique where it can further reduce the noise of the digital mammograms while preserving the contrast and brightness. A total of six digital mammograms were retrieved from Mammographic Image Analysis Society (MIAS) open-source database. The performance of FC-CLAHE-ADF was compared to Recursive Mean-Separate Histogram Equalization (RMSHE), Fast Discrete Curvelet Transform via Unequally Spaced Fast Fourier Transform (FDCT-USFFT), and FC-CLAHE only. In summary, this novel FC-CLAHE-ADF has provided the most superior results, among other selected enhancement techniques. The resulting images have been able to demonstrate breast lesions better.

Convolutional neural network for automated mass segmentation in mammography

BackgroundAutomatic segmentation and localization of lesions in mammogram (MG) images are challenging even with employing advanced methods such as deep learning (DL) methods. We developed a new model based on the architecture of the semantic segmentation U-Net model to precisely segment mass lesions in MG images. The proposed end-to-end convolutional neural network (CNN) based model extracts contextual information by combining low-level and high-level features. We trained the proposed model using huge publicly available databases, (CBIS-DDSM, BCDR-01, and INbreast), and a private database from the University of Connecticut Health Center (UCHC).ResultsWe compared the performance of the proposed model with those of the state-of-the-art DL models including the fully convolutional network (FCN), SegNet, Dilated-Net, original U-Net, and Faster R-CNN models and the conventional region growing (RG) method. The proposed Vanilla U-Net model outperforms the Faster R-CNN model significantly in terms of the runtime and the Intersection over Union metric (IOU). Training with digitized film-based and fully digitized MG images, the proposed Vanilla U-Net model achieves a mean test accuracy of 92.6%. The proposed model achieves a mean Dice coefficient index (DI) of 0.951 and a mean IOU of 0.909 that show how close the output segments are to the corresponding lesions in the ground truth maps. Data augmentation has been very effective in our experiments resulting in an increase in the mean DI and the mean IOU from 0.922 to 0.951 and 0.856 to 0.909, respectively.ConclusionsThe proposed Vanilla U-Net based model can be used for precise segmentation of masses in MG images. This is because the segmentation process incorporates more multi-scale spatial context, and captures more local and global context to predict a precise pixel-wise segmentation map of an input full MG image. These detected maps can help radiologists in differentiating benign and malignant lesions depend on the lesion shapes. We show that using transfer learning, introducing augmentation, and modifying the architecture of the original model results in better performance in terms of the mean accuracy, the mean DI, and the mean IOU in detecting mass lesion compared to the other DL and the conventional models.

Deep Learning Pre-training Strategy for Mammogram Image Classification: an Evaluation Study.

In this work, we assess how pre-training strategy affects deep learning performance for the task of distinguishing false-recall from malignancy and normal (benign) findings in digital mammography images. A cohort of 1303 breast cancer screening patients (4935 digital mammogram images in total) was retrospectively analyzed as the target dataset for this study. We assessed six different convolutional neural network model structures utilizing four different imaging datasets (total > 1.4 million images (including ImageNet); medical images different in terms of scale, modality, organ, and source) for pre-training on six classification tasks to assess how the performance of CNN models varies based on training strategy. Representative pre-training strategies included transfer learning with medical and non-medical datasets, layer freezing, varied network structure, and multi-view input for both binary and triple-class classification of mammogram images. The area under the receiver operating characteristic curve (AUC) was used as the model performance metric. The best performing model out of all experimental settings was an AlexNet model incrementally pre-trained on ImageNet and a large Breast Density dataset. The AUC for the six classification tasks using this model ranged from 0.68 to 0.77. In the case of distinguishing recalled-benign mammograms from others, four out of five pre-training strategies tested produced significant performance differences from the baseline model. This study suggests that pre-training strategy influences significant performance differences, especially in the case of distinguishing recalled- benign from malignant and benign screening patients.