Interpretable Sepsis Infection Prediction Based On Kolmogorov-Arnold Neural Network
Code: G-1831
Authors: Fatemeh Torabi Konjin * ℗, Sarina Azimian, Rezvan Gholami
Schedule: Not Scheduled!
Tag: Biomedical Signal Processing
Download: Download Poster
Abstract:
Abstract
Background and aims: Sepsis is a type of illness that occurs as a result of the body's incorrect response to infection, characterized by organ weakness and sometimes shock, which can result in death for patients. Some reports indicate that this disease has a higher mortality rate compared to breast and prostate cancer worldwide .Any efforts aimed at the early diagnosis of infection and treatment can increase the chances of survival for patients. Given the emergence of computer-based decision support systems powered by artificial intelligence in healthcare facilities and hospitals in recent years, the objective of this study is to evaluate the Kolmogorov-Arnold neural network method in predicting sepsis infection. Method: This study utilized clinical and laboratory data from 65,321 patients who visited a hospital in Shanghai, China, between the years 2016 and 2021. After preprocessing which includes managing missing data, balancing using under sampling Technique, identifying and removing outlier and normalizing the data, the implementation of the Kolmogorov-Arnold neural network was carried out in the Google Colab environment using the Python programming language. Results: After training the model and using the Adam optimizer to update the model parameters ,To assess the effectiveness of this model, the accuracy metric has been used. The accuracy of this model with learnable activation functions along edges, has been reported to be 91.5 %.This takes 2.58 minutes to train. Conclusion: The results of this study indicate that this method has a higher accuracy in diagnosing sepsis infections compared to conventional deep learning methods such as convolutional neural networks and long short-term memory networks. This method enhances physicians' understanding of the reasoning behind predictions, thereby improving clinical decision-making, by providing a transparent model with an interpretable architecture.
Keywords
Kolmogorov-Arnold Neural Network, Interpretability
