2-HDCNN: A two-tier hybrid dual convolution neural network feature fusion approach for diagnosing malignant melanoma.

Melanoma is a fatal form of skin cancer, which causes excess skin cell growth in the body. The objective of this work is to develop a two-tier hybrid dual convolution neural network (2-HDCNN) feature fusion approach for malignant melanoma prediction. The first-tier baseline Convolutional Neural Network (CNN) extracts the hard to classify samples based on the confidence factor (class probability variance score) and generates a Baseline Segregated Dataset (BSD). The BSD is then preprocessed using hair removal and data augmentation techniques. The preprocessed BSD is trained with the second-tier CNN that yields the bottleneck features. These features are then combined with the derived features from the ABCD (Asymmetry, Border, Color and Diameter) medical rule to improve classification accuracy. The generated hybrid fused features are fed to different classifiers like Gradient boosting classifiers, Bagging classifiers, XGBoost classifiers, Decision trees, Support Vector Machine, Logistic regression and Multi-layer perceptron. For performance assessment, the proposed framework is trained on the ISIC 2018 dataset. The experimental results prove that the presented 2-HDCNN feature fusion approach has reached an accuracy of 92.15%, precision of 96.96%, specificity of 96.8%, sensitivity of 86.48%, and AUC (Area Under Curve) value of 0.96 for diagnosing malignant melanoma.

Computer-Aided Diagnosis system for diagnosis of pulmonary emphysema using bio-inspired algorithms.

Pulmonary emphysema is a condition characterized by the destruction and permanent enlargement of the alveoli of the lungs. The destruction of gas-exchanging alveoli causes shortness of breath followed by a chronic cough and sputum production. A Computer-Aided Diagnosis (CAD) framework for diagnosing pulmonary emphysema from chest Computed Tomography (CT) slices has been designed and implemented in this study. The process of implementing the CAD framework includes segmenting the lung tissues and extracting the regions of interest (ROIs) using the Spatial Intuitionistic Fuzzy C-Means clustering algorithm. The ROIs that were considered in this work were emphysematous lesions - namely, centrilobular, paraseptal, and bullae that were labelled by an expert radiologist. The shape, texture, and run-length features were extracted from each ROI. A wrapper approach that employed four bio-inspired algorithms - namely, Moth-Flame Optimization (MFO), Firefly Optimization (FFO), Artificial Bee Colony Optimization, and Ant Colony Optimization - with the accuracy of the support vector machine classifier as the fitness function was used to select the optimal feature subset. The selected features of each bio-inspired algorithm were trained independently using the Extreme Learning Machine classifier based on the tenfold cross-validation technique. The framework was tested on real-time and public emphysema datasets to perform binary classification of lung CT slices of patients with and without the presence of emphysema. The framework that used MFO and FFO for feature selection produced superior results regarding accuracy, precision, recall, and specificity for the real-time dataset and the public dataset, respectively, when compared to the other bio-inspired algorithms.

Feature selection using ant colony optimization with tandem-run recruitment to diagnose bronchitis from CT scan images.

BACKGROUND AND OBJECTIVES: Computer-aided diagnosis (CAD) plays a vital role in the routine clinical activity for the detection of lung disorders using computed tomography (CT) images. It serves as a source of second opinion that radiologists may consider in order to interpret CT images. In this work, the purpose of CAD is to improve the diagnostic accuracy of pulmonary bronchitis from CT images of the lung. METHODS: Left and right lung fields are segmented using optimal thresholding from the lung CT images. Texture and shape features are extracted from the pathology bearing regions. A hybrid feature selection approach based on ant colony optimization (ACO) combining cosine similarity and support vector machine (SVM) classifier is used to select relevant features. Additionally, tandem run recruitment strategy is included in the selection activity to choose the promising features. The SVM classifier is trained using the selected features and the performance of the trained classifier is evaluated using trivial performance evaluation measures. RESULTS: The training and testing datasets used in building the classifier model are disjoint and contains 200 CT slices affected with bronchitis, 50 normal slices and 300 slices with cancer. Out of 100 features extracted from each CT slice, a subset of 60 features is used for classification. ACO with tandem run strategy yielded 81.66% of accuracy whereas ACO without tandem run yielded an accuracy of 77.52%. When all the features are used for classifier training without feature selection algorithm, an accuracy of 75.14% is achieved. CONCLUSION: From the results, it is inferred that identifying relevant features to train the classifier has a definite impact on the classifier performance.

A Swarm Optimization approach for clinical knowledge mining.

BACKGROUND AND OBJECTIVES: Rule-based classification is a typical data mining task that is being used in several medical diagnosis and decision support systems. The rules stored in the rule base have an impact on classification efficiency. Rule sets that are extracted with data mining tools and techniques are optimized using heuristic or meta-heuristic approaches in order to improve the quality of the rule base. In this work, a meta-heuristic approach called Wind-driven Swarm Optimization (WSO) is used. The uniqueness of this work lies in the biological inspiration that underlies the algorithm. METHODS: WSO uses Jval, a new metric, to evaluate the efficiency of a rule-based classifier. Rules are extracted from decision trees. WSO is used to obtain different permutations and combinations of rules whereby the optimal ruleset that satisfies the requirement of the developer is used for predicting the test data. The performance of various extensions of decision trees, namely, RIPPER, PART, FURIA and Decision Tables are analyzed. The efficiency of WSO is also compared with the traditional Particle Swarm Optimization. RESULTS: Experiments were carried out with six benchmark medical datasets. The traditional C4.5 algorithm yields 62.89% accuracy with 43 rules for liver disorders dataset where as WSO yields 64.60% with 19 rules. For Heart disease dataset, C4.5 is 68.64% accurate with 98 rules where as WSO is 77.8% accurate with 34 rules. The normalized standard deviation for accuracy of PSO and WSO are 0.5921 and 0.5846 respectively. CONCLUSION: WSO provides accurate and concise rulesets. PSO yields results similar to that of WSO but the novelty of WSO lies in its biological motivation and it is customization for rule base optimization. The trade-off between the prediction accuracy and the size of the rule base is optimized during the design and development of rule-based clinical decision support system. The efficiency of a decision support system relies on the content of the rule base and classification accuracy.