Publications
Implementation of Non-Contact Bed Embedded Ballistocardiogram Signal Measurement and Valvular Disease Detection
Authors: Hafiz, M.A., Hashem, A.M., Khan, A.A.S., Rashid, M.H., Faruqui, M.A.K.
Published in: International Journal of Medical Engineering and Informatics
Abstract: In a traditional system, ECG leads are connected to the patient's chest to detect the electrical performance of the heart which create long term discomfort for the patient. As ballistocardiogram (BCG) and valvular diseases are both mechanical phenomena, we conjectured that valvular disease could be diagnosed from non-contact BCG measurement. In this paper, we proposed a non-contact way to determine the valvular diseases of the heart which is favourable for long term observation of the patient. We classified the data using artificial neural network (ANN) and support vector machine (SVM). We collected data from normal persons and persons affected by mitral and pulmonary valve stenosis. We compared the result using overall accuracy, misclassification rate and fitness. We got the highest test accuracy of 79.12% for SVM technique for decomposition level 1. As this technique is completely new and advantageous, it can lead to a new research area of valvular disease detection.
Volume: 13, Issue: 4, Pages: 289–296, Year: 2021
DOI: 10.1504/IJMEI.2021.115970
Diagnosis of Malignant Melanoma using Color and Textural Features from Dermoscopic Images
Authors: Shahrin Akter, Joynob Binte Ahmed, M. A. Hafiz, Nusrat Jahan
Published in: International Conference on Big Data, IoT and Machine Learning (BIM 2021)
Abstract: The most common cancer in the world is skin cancer. Among various types of skin cancer, melanoma is less common but severe because of its quick-spreading-out property. Melanoma patients, fortunately, can be cured if detected and treated early. We have proposed a feed-forward back-propagation neural network-based technique in this paper using color and textural features for the early diagnosis of melanoma. This technique consists of image acquisition, pre-processing, lesion segmentation, features extraction, and classification of the lesion. We have assembled texture features with color variation to have greater detection accuracy. We have chosen a neural network-based classifier because it is both easy and capable of reducing error. As far as we know, our proposed method using color and textural features shows more accuracy than others. We have got classification accuracy of 97%. It also gives a sensitivity of 92.5% and a specificity of 98.12%.
Date: September 23-25, 2021