Application of Artificial Intelligence to Advance Individualized Diagnosis and Treatment in Emergency and Critical Care Medicine

Jie Yang,Bo Zhang,Xiaocong Jiang,Jiajie Huang,Yucai Hong,Hongying Ni,Zhongheng Zhang

DOI: https://doi.org/10.3390/diagnostics14070687

IF: 3.6

2024-03-26

Diagnostics

Abstract:Emergency and critical illnesses refer to severe diseases or conditions characterized by rapid changes in health that may endanger life within a short period [...]

medicine, general & internal

What problem does this paper attempt to address?

The main purpose of this paper is to explore the application of Artificial Intelligence (AI) in emergency and critical care medicine to achieve advancements in personalized diagnosis and treatment. Specifically, the paper covers the following aspects:

1. **Application of Explainable AI (XAI) in Diagnosis**:
- Aygun et al. developed an XAI-based interpretable predictive model for sepsis prediction and identification of important biomarkers. The study results indicate that factors such as age, increased respiratory rate, and decreased procalcitonin levels increase the risk of sepsis.

2. **Diagnostic Method Selection for Acute Appendicitis**:
- Tuncyurek et al. evaluated the effectiveness of the American College of Radiology (ACR) appropriateness criteria in diagnosing acute appendicitis in children, aiming to optimize the selection of diagnostic methods through multivariable decision criteria and reduce unnecessary surgeries.

3. **Study of Risk Factors for Acute Kidney Injury (AKI)**:
- Researchers screened the MIMIC-IV database and identified serum creatinine, chloride, and magnesium as the three main indicators that need to be closely monitored to better understand the clinical management of AKI patients.

4. **Early Detection of Ventilator-Associated Pneumonia (VAP)**:
- Rambaud et al. developed a clinical prediction algorithm using data from a Canadian tertiary pediatric hospital, achieving highly sensitive predictions for VAP, aiding in early identification and intervention.

5. **Prognostic Assessment of Traumatic Brain Injury (TBI) and Burn Patients**:
- Tu et al. and Yeh et al. applied machine learning models in TBI and burn patients, respectively, improving the accuracy of predicting mortality risk and other adverse outcomes.

6. **Prediction of Mortality Risk in ICU Patients**:
- Researchers used data from the MIMIC-IV database to build various machine learning models and found that the XGBoost method outperformed traditional models in predicting patient mortality risk.

7. **Prediction of ICU Needs for COVID-19 Patients**:
- Islam et al. analyzed data collected from the pulmonology department of a state hospital in Moscow and proposed a stacked machine learning model to assess the risk of ICU admission for COVID-19 and non-COVID patients.

8. **Neonatal Pain Assessment**:
- Zhu et al. developed a video-based automated neonatal pain assessment system, validated its effectiveness with actual clinical data, and demonstrated its feasibility in remote neonatal pain assessment.

In summary, these studies demonstrate the transformative impact of AI and machine learning in medical diagnosis and prognosis, advancing precision medicine, potentially improving patient outcomes, and optimizing healthcare resource allocation. However, the studies also highlight the issue of insufficient model generalization, emphasizing the need for further validation through multicenter randomized controlled trials (RCTs).