Labelled Feature Dimensionality Reduction for Liver Disorder Classification in CT Liver Images

Features play a big role in computer vision. For a variety of image categorization tasks using features, structures based on previously learned features can be used. Using dimensionality reduction techniques, the dimensionality of the inner layers is reduced. Convolution Neural Networks (CNNs) were originally designed to identify the most essential parts of a image using efficient techniques, but these qualities are now routinely learned by different layers of CNNs in the same way they were originally designed. When a neural network has bidirectional long-short term memory (Bi-LSTM), information can be stored in both directions like forward and backward. When using a standard LSTM, input flow in one direction: backwards or forwards. However, with bi-directional, the ability to keep both the future and the past knowledge in mind at the same time is possible. In this research, trained CNN features integrated with BiLSTM are used to build a computer vision system that is universally applicable. For a variety of image categorization tasks, structures based on previously learned features can be used. Feature dimensionality reduction strategies are used to reduce the high dimensionality of the image inner layer features. At the same time as gathering critical parameters under one influential parameter, feature extraction reduces many dimensions simultaneously. To make accurate predictions about liver tumours these days, doctors must use better detection mechanisms. A liver tumour may be predicted by experienced practitioners through experience and imaging, scan, etc. but system-oriented assistance is needed to make a final conclusion based on the symptoms of patients. Classification and disease analysis are made possible because to the comprehensive range of features. In this research, a Labeled Feature Dimensionality Reduction Model using Bi-LSTM (LFDR-BiLSTM) model is proposed that effectively reduces the extracted features for liver disorder classification. The proposed model is contrasted with the existing model and the results represent that the proposed model performance is effective.