Efficient sepsis detection using deep learning and residual convolutional networks.

Sepsis is a life-threatening complication caused by infection that leads to extensive tissue damage. If not treated promptly, it can become fatal. Early identification and diagnosis of sepsis are critical to improving patient outcomes. Although recent technological advancements have aided sepsis detection, challenges remain in timely diagnosis using standard clinical practices. In this article, we present a new deep learning model to detect the occurrence of sepsis and the African vulture optimization algorithm (AVOA) to enhance the model performance. The system comprises four crucial steps: First, the enhanced convolutional learning framework (ECLF) with atrous convolutional and multi-level strategies that aim to learn high-level features from the nonlinear mapping of the medical data. Second is the spatio-channel attention network (SCAN), which has a neural architecture designed to focus on significant regions, such as spatial and channel regions, but not restricted to them. Third is the hierarchical dilated convolutional block (HDCB), which utilises a stacked dilated deep convolutional architecture for spatial feature context retrieval. Last is the residual path convolutional chain (RPCC), which uses a multi-residual convolutional approach for feature propagation, preserving important information. The sepsis detection model we bring forth involves many components, as mentioned above, and thus achieves a higher accuracy for timely intervention during sepsis. The combination of AVOA into the model ensures that it is robust and easily transferable, delivering high performance for adaptation to complicated structures inside medical datasets. The proposed model was evaluated on a clinical dataset and achieved outstanding performance, with an accuracy of 99.4%, precision of 98%, recall of 99.2%, F1-score of 99.0%, and an area under the curve (AUC) of 0.998. These results demonstrate the model's superior ability to detect sepsis accurately and reliably, outperforming traditional clinical scoring methods and conventional machine learning approaches.