Alicia kim's blog

Min-Kyung Jung, Tae Hoon Roh, Hakseung Kim, Eun Jin Ha, Dukyong Yoon, Chan Min Park, Se-Hyuk Kim, Namkyu You, Dong-Joo Kim (Korea University)

Expert Systems with Applications (2024)

Abstract:

• Develops an explainable neural network model to forecast physiological responses after osmotic therapy in traumatic brain injury (TBI) patients.
• Data from arterial blood pressure, intracranial pressure – 107 TBI patients treated with mannitol or hypertonic saline.
• Six prediction models were built using the LTSF-NLinear time series forecasting approach for 1-hour predictions.
• The model’s predictions were evaluated for both physiological signal and event prediction.

Findings:

• The forecasting models achieved high performance: R² > 0.8 for physiological signals and accuracy > 0.9 for predicting adverse events.
• Including both mannitol and hypertonic saline treatments allowed the model to generalize across two types of osmotic therapy.
• Provided insight into which recent physiological signal values most influence model predictions.
• Local explanations → how individual patient states affected predicted responses.

Conclusion:

• Neural network forecasting – predict short-term responses to osmotic therapy in TBI clinical settings.
• Model – Quantitative guidance for clinicians to support decisions on therapy maintenance and transition.
• High predictive accuracy and interpretability make this approach promising for improving prognosis and clinical workflow in neurological intensive care.
• Both global and local explanations → physiological response patterns post-treatment.