Human-in-the-Loop Predictive Analytics Using Statistical Learning.

Area of Science:

• Medical Technology
• Artificial Intelligence in Healthcare
• Neurological Prognosis

Background:

• Current coma prediction relies solely on device data, often missing crucial human input for timely decisions.
• Advancements in automation and AI offer potential for improved early disease diagnosis in medicine.
• Coma presents a critical and challenging diagnostic problem in medical research.

Purpose of the Study:

• To propose a healthcare framework integrating artificial intelligence within a human-in-the-loop cyber-physical system.
• To design and evaluate a model for the early prognosis of coma utilizing electroencephalogram (EEG) datasets.
• To compare the efficacy of statistical learning algorithms against deep learning models for coma prediction.

Main Methods:

• Development of a human-in-the-loop cyber-physical system with a response loop for decision interpretation.
• Application of the autoregressive integrated moving average (ARIMA) statistical learning algorithm to EEG data.
• Comparative analysis with artificial neural networks (ANNs) and long short-term memory (LSTM) models.

Main Results:

• The autoregressive integrated moving average model demonstrated superior performance in early coma prognosis compared to ANNs and LSTMs.
• The proposed model achieved the lowest error values, as measured by root mean squared error (RMSE), mean absolute error (MAE), and mean squared error (MSE).
• The study utilized a publicly available dataset from the PhysioNet open-source community.

Conclusions:

• The proposed human-in-the-loop AI framework, particularly using ARIMA, shows significant promise for early coma detection.
• Integrating human intention with AI analysis of biological signals enhances diagnostic accuracy in critical conditions.
• The findings highlight the effectiveness of statistical learning models in analyzing EEG data for neurological prognoses.