Computer-aided classification of mammographic masses using visually sensitive image features

Abstract

Purpose To develop a new computer-aided diagnosis (CAD) scheme that computes visually sensitive image features routinely used by radiologists to develop a machine learning classifier and distinguish between the malignant and benign breast masses detected from digital mammograms. Methods An image dataset including 301 breast masses was retrospectively selected. From each segmented mass region, we computed image features that mimic five categories of visually sensitive features routinely used by radiologists in reading mammograms. We then selected five optimal features in the five feature categories and applied logistic regression models for classification. A new CAD interface was also designed to show lesion segmentation, computed feature values and classification score. Results Areas under ROC curves (AUC) were 0.786±0.026 and 0.758±0.027 when to classify mass regions depicting on two view images, respectively. By fusing classification scores computed from two regions, AUC increased to 0.806±0.025. Conclusion This study demonstrated a new approach to develop CAD scheme based on 5 visually sensitive image features. Combining with a “visual aid” interface, CAD results may be much more easily explainable to the observers and increase their confidence to consider CAD generated classification results than using other conventional CAD approaches, which involve many complicated and visually insensitive texture features.

Prediction of breast cancer risk using a machine learning approach embedded with a locality preserving projection algorithm

“…This result is quite interesting and may be worth further investigation. Using a local instance based learning method (i.e., a KNN algorithm) can provide great flexibility to develop a new machine learning based imaging marker or prediction mode because it will be relatively easy to periodically add new image data to increase size and diversity of the reference database for the instance-based learning model, without a complicated retraining to produce a global optimization function, which is required by all other “eager” learning methods (Park et al 2007, Wang et al 2017).…”

Developing global image feature analysis models to predict cancer risk and prognosis

“…The scheme initially computes 59 global mammographic image features, followed by applying a particle swarm optimization algorithm to search for optimal features and training a support vector machine model to predict the likelihood of malignancy. When using a relatively small dataset involving 134 malignant and 141 benign cases, the model yields a performance of AUC = 0.79 ± 0.07 [58], which is highly comparable to the performance of applying tumor-based CAD schemes in classifying malignant and benign tumors [26].…”

“…Although developing deep learning models can avoid tumor segmentation, it requires "big data" (availability of large training datasets). Thus, besides working to investigate how to optimally apply the deep learning method to develop robust CAD schemes using the small image datasets [26][27][28], we also investigate a different conventional machine learning approach that uses global image features computed from the entire imaged organs (i.e., breast, lung, and abdominal region) to train prediction models without suspicious region or tumor segmentation (as used in the conventional tumor-based schemes) or predefine the regions of interests with a fixed size (as used in many deep learning-based schemes). The new global image feature analysis-based models can be either implemented to build new case-based CAD schemes or fused with the existing tumor-based CAD schemes.…”

Methods Used in Computer-Aided Diagnosis for Breast Cancer Detection Using Mammograms: A Review

“…Area under ROC Binary decision tree [210] 0.90 Linear classifier [210] 0.90 PCA-LS SVM [211] 0.94 ANN [212] 0.88 Multiple expert system [213] 0.79 Texture measure with ANN [214] 0.87 Multiresolution texture analysis [215] 0.86 Subregion Hotelling observers [216] 0.94 Logistic regression [217] 0.81 KNN [218] 0.82 NB [219] 0.56 DL [190] 0.96 Genetic algorithms with SVM [220] 0.97 through a combining schema. From an implementation point of view, combination topologies can be categorized into multiple, conditional, hierarchical, or hybrid topologies [199].…”