Hybrid fractional groupers and moray eels driven deep learning for pneumonia detection using multi-modal data in federated learning.

Authors

Abstract

Pneumonia is a severe lung infection triggered by various viral pathogens. Detecting and diagnosing pneumonia using clinical images is challenging because its visual features often resemble those of other pulmonary conditions. As a result, existing approaches for pneumonia prediction frequently face struggles in achieving high accuracy and also face issues in protecting the security of the medical data. Thus, this paper presents a novel model named Fractional Groupers and Moray Eels optimization_Euclidean, Expectation loss (FGMEO_EESHLossNet) for pneumonia detection utilizing multi-modal data within a Federated Learning (FL) framework. The process begins with local training at each node using its respective data. Subsequently, local data are updated, and model aggregation occurs on the server. Then, the global model is downloaded at each node, and the training process is updated using the downloaded global and local models; this is iterative at every epoch. The pneumonia detection is performed in a training model, such that the multi-modal dataset used for training includes Computed Tomography (CT), chest X-ray, and spectrogram images. Initially, chest X-ray images are denoised with an Alpha trimmed mean filter, and the contrast of the image is enhanced through gamma correction. Then, segmentation of the lung lobe from chest X-ray images is performed using a Deep Recursive Residual Network (DRRN). Here, contrast-enhanced images are segmented to isolate affected areas using the DRRN. Following segmentation, image augmentation and feature extraction are conducted. Similarly, the same process is applied to CT images, and the de-noising and contrast enhancement steps are applied to spectrogram images. Moreover, the outputs from all three modalities are fed into the Shepard Convolution Pyramid Dilated Network for pneumonia detection, and the network's layers are modified using the EESHLossNet loss function. Here, the EESHLossNet is trained by FGMEO, which is the combination of the Fractional Calculus (FC) concept with Groupers and Moray Eels Optimization (GMEO). Additionally, the devised model has achieved Loss function, Mean Squared Error (MSE), accuracy, True Positive Rate (TPR), True Negative Rate (TNR), Precision, F1-Score, False Negative Rate (FNR), and False Positive Rate (FPR) as 0.075, 0.087, 0.925, 0.915, 0.933, 0.933, 0.924, 0.085, and 0.067, for time stamp of 100s.

Tags

Topics