In this study, we investigated the feasibility of using a combination of radiomic features and clinical parameters to predict the duration of IMV exceeding 7 days in TBI patients. In our research findings, we employed five different machine learning algorithms to construct a radiomics model and selected the best model, which achieved an AUC of 0.8496 in predicting TBI patients requiring IMV for more than 7 days. Furthermore, the combination of R-score and clinical parameters improved the prediction of TBI patients requiring IMV for more than 7 days, with an AUC of 0.8959. These results demonstrated the predictive ability of combining radiomics with clinical parameters for the duration of IMV exceeding 7 days in TBI patients.
Radiomic features are quantitative data obtained from medical images, including first-order statistics as well as features based on shape and texture. Mathematically defined, radiomic features represent the image characteristics of a ROI, many of which cannot be discerned through visual interpretation. It has been shown that radiomic features can provide insights into lesion characteristics and can be further modeled using machine learning algorithms based on selected features for diagnostic, prognostic, or predictive purposes.16 Radiomics analysis has great potential in hemorrhagic diseases and is helpful for the etiological diagnosis of intracerebral hemorrhage and for detecting hematoma expansion in cases of spontaneous intracerebral hemorrhage.17 However, it is unclear whether radiomics features can predict the duration of IMV exceeding 7 days. Our study indicated the value of radiomics signatures in detecting the duration of IMV exceeding 7 days.
Since radiomic features only consider the characteristics of intracranial hemorrhage on CT images, it is crucial to incorporate other important clinical parameters to predict adverse outcomes. In this study, several routinely available clinical parameters were selected, and ultimately, a radiomic-clinical combined model was established to predict adverse outcomes. Univariate analysis revealed that lower GCS, higher ISS, lower PAB count, higher NEUT count were significantly associated with TBI patients requiring IMV exceeding 7 days. Other parameters may also be potential risk factors, but their significance was not observed in this study. The GCS provides an objective assessment of patients’consciousness. A higher score indicates a better consciousness, while a lower GCS primarily indicates severe brain injury and has been observed to be associated with poor functional recovery.18,19 In a study conducted in Taiwan, China, it was reported that a lower GCS was a predictive factor for failed weaning from prolonged IMV.20 This finding confirms the correlation between a lower GCS and a longer duration of IMV in patients. The ISS is commonly used to measure the severity of a patient's injuries. Any patient with an ISS > 15 is considered to have a severe injury. In this study, the average ISS for both groups of TBI patients was greater than 15 (Table 1). In mechanically ventilated trauma patients, a higher ISS is often associated with a longer duration of mechanical ventilation.21–23 Serum ALB is a nutritional marker that can also influence the duration of IMV in TBI patients. Low levels of ALB are a useful prognostic marker as they result from increased leakage of albumin from capillaries into the interstitial fluid due to systemic inflammatory response syndrome, which has been observed in ICU patients.24 Michael et al. also found that serum ALB on the first day of admission significantly affects the duration of IMV, with lower levels of serum ALB being significantly associated with longer IMV duration.25 Trauma can lead to the release of various inflammatory mediators and trigger inflammation.26 During the acute phase of inflammation, various inflammatory cells accumulate and infiltrate, inducing systemic inflammatory response syndrome, and further progression of inflammatory cells can lead to multiple organ dysfunction syndrome.27 Moreover, the increase in inflammatory activators such as NEUT can cause vascular dysfunction in the early stages of injury, directly leading to disruption of the blood-brain barrier,28 thereby exacerbating the progression of the condition. In summary, clinical parameters also play a significant role in predicting the duration of IMV in TBI patients.
In previous radiomics studies, there have been no reports on the association between the duration of invasive mechanical ventilation and TBI patients. This study confirms the correlation between the two. The significance of this research lies in the following aspects: (1) Improving treatment effectiveness: By predicting the duration of IMV, clinicians can better develop treatment plans, provide timely rehabilitation interventions, reduce mortality and disability rates, and improve treatment effectiveness; (2) Reducing healthcare resource waste: Through this prediction, healthcare resources can be utilized efficiently, avoiding unnecessary treatments and examinations, thereby reducing wastage of healthcare resources; (3) Advancement of medical knowledge: The results of this research methodology may provide new insights and algorithms for the medical field, contributing to advancements in medical knowledge. Overall, this study has important implications for improving patient outcomes, optimizing resource allocation, and advancing medical research.
Some limitations are noted in this study. Firstly, it is a single-center retrospective study, lacking external validation with data from other centers, and due to the limited number of cases, we were unable to validate using an independent dataset. Secondly, retrospective analysis includes biases related to patient selection. Lastly, currently incorporating the manual segmentation process into clinical workflow is time-consuming and impractical. We attempted to apply automated segmentation tools to our dataset; however, the segmentation results for traumatic IPH were not satisfactory and require extensive adjustments. Compared to spontaneous IPH, the diversity and lower imaging contrast of traumatic IPH lesions make automated segmentation more challenging, mainly due to their proximity to the bony structures and extra-axial hemorrhage. However, this field is rapidly evolving. By utilizing fine-tuned automated segmentation tools for traumatic IPH, timely processing of a large volume of images for radiomics analysis can be achieved in the near future.