Towards Personalised Contrast Injection: Artificial-Intelligence-Derived Body Composition and Liver Enhancement in Computed Tomography

The association of radiologic body composition parameters with clinical outcomes in level-1 trauma patients

PURPOSE

The present study aims to assess whether CT-derived muscle mass, muscle density, and visceral fat mass are associated with in-hospital complications and clinical outcome in level-1 trauma patients.

METHODS

A retrospective cohort study was conducted on adult patients admitted to the University Medical Center Utrecht following a trauma between January 1 and December 31, 2017. Trauma patients aged 16 years or older without severe neurological injuries, who underwent a CT that included the abdomen within 7 days of admission, were included. An artificial intelligence (AI) algorithm was used to retrieve muscle areas to calculate the psoas muscle index and to retrieve psoas muscle radiation attenuation and visceral fat (VF) area from axial CT images. Multivariable logistic and linear regression analyses were performed to assess associations between body composition parameters and outcomes.

RESULTS

A total of 404 patients were included for analysis. The median age was 49 years (interquartile range [IQR] 30-64), and 66.6% were male. Severe comorbidities (ASA 3-4) were seen in 10.9%, and the median ISS was 9 (IQR 5-14). Psoas muscle index was not independently associated with complications, but it was associated with ICU admission (odds ratio [OR] 0.79, 95% confidence interval [CI] 0.65-0.95), and an unfavorable Glasgow Outcome Scale (GOS) score at discharge (OR 0.62, 95% CI 0.45-0.85). Psoas muscle radiation attenuation was independently associated with the development of any complication (OR 0.60, 95% CI 0.42-0.85), pneumonia (OR 0.63, 95% CI 0.41-0.96), and delirium (OR 0.49, 95% CI 0.28-0.87). VF was associated with developing a delirium (OR 1.95, 95% CI 1.12-3.41).

CONCLUSION

In level-1 trauma patients without severe neurological injuries, automatically derived body composition parameters are able to independently predict an increased risk of specific complications and other poor outcomes.

Artificial intelligence and body composition

AIMS

Although obesity is associated with chronic disease, a large section of the population with high BMI does not have an increased risk of metabolic disease. Increased visceral adiposity and sarcopenia are also risk factors for metabolic disease in people with normal BMI. Artificial Intelligence (AI) techniques can help assess and analyze body composition parameters for predicting cardiometabolic health. The purpose of the study was to systematically explore literature involving AI techniques for body composition assessment and observe general trends.

METHODS

We searched the following databases: Embase, Web of Science, and PubMed. There was a total of 354 search results. After removing duplicates, irrelevant studies, and reviews(a total of 303), 51 studies were included in the systematic review.

RESULTS

AI techniques have been studied for body composition analysis in the context of diabetes mellitus, hypertension, cancer and many specialized diseases. Imaging techniques employed for AI methods include CT (Computerized Tomography), MRI (Magnetic Resonance Imaging), ultrasonography, plethysmography, and EKG(Electrocardiogram). Automatic segmentation of body composition by deep learning with convolutional networks has helped determine and quantify muscle mass. Limitations include heterogeneity of study populations, inherent bias in sampling, and lack of generalizability. Different bias mitigation strategies should be evaluated to address these problems and improve the applicability of AI to body composition analysis.

CONCLUSIONS

AI assisted measurement of body composition might assist in improved cardiovascular risk stratification when applied in the appropriate clinical context.

Individualized Contrast Media Application Based on Body Weight and Contrast Enhancement in Computed Tomography of Livers without Steatosis

This study analyzes the homogeneity in liver attenuation of a body-weight-based protocol compared to a semi-fixed protocol. Patients undergoing abdominal multiphase computed tomography received 0.500 g of iodine (gI) per kilogram of body weight. Liver attenuation and enhancement were determined using regions of interest on scans in the pre-contrast and portal venous phases. The outcomes were analyzed for interpatient uniformity in weight groups. The subjective image quality was scored using a four-point Likert scale (excellent, good, moderate, and nondiagnostic). A total of 80 patients were included (56.3% male, 64 years, 78.0 kg) and were compared to 80 propensity-score-matched patients (62.5% male, 63 years, 81.7 kg). The liver attenuation values for different weight groups of the TBW-based protocol were not significantly different (p = 0.331): 109.1 ± 13.8 HU (≤70 kg), 104.6 ± 9.70 HU (70−90 kg), and 105.1 ± 11.6 HU (≥90 kg). For the semi-fixed protocol, there was a significant difference between the weight groups (p < 0.001): 121.1 ± 12.1 HU (≤70 kg), 108.9 ± 11.0 HU (70−90 kg), and 105.0 ± 9.8 HU (≥90 kg). For the TBW-based protocol, the enhancement was not significantly different between the weight groups (p = 0.064): 46.2 ± 15.1 HU (≤70 kg), 59.3 ± 6.8 HU (70−90 kg), and 52.1 ± 11.7 HU (≥90 kg). Additionally, for the semi-fixed protocol, the enhancement was not significantly different between the weight groups (p = 0.069): 59.4 ± 11.0 HU (≤70 kg), 53.0 ± 10.3 HU (70−90 kg), and 52.4 ± 7.5 HU (≥90 kg). The mean administered amount of iodine per kilogram was less for the TBW-based protocol compared to the semi-fixed protocol: 0.499 ± 0.012 and 0.528 ± 0.079, respectively (p = 0.002). Of the TBW-based protocol, 17.5% of the scans scored excellent enhancement quality, 76.3% good, and 6.3% moderate. Of the semi-fixed protocol, 70.0% scored excellent quality, 21.3% scored good, and 8.8% scored moderate. In conclusion, the TBW-based protocol increased the interpatient uniformity of liver attenuation but not the enhancement in the portal venous phase compared to the semi-fixed protocol, using an overall lower amount of contrast media and maintaining good subjective image quality.

Deep learning body-composition analysis of clinically acquired CT-scans estimates creatinine excretion with high accuracy in patients and healthy individuals

Assessment of daily creatinine production and excretion plays a crucial role in the estimation of renal function. Creatinine excretion is estimated by creatinine excretion equations and implicitly in eGFR equations like MDRD and CKD-EPI. These equations are however unreliable in patients with aberrant body composition. In this study we developed and validated equations estimating creatinine production using deep learning body-composition analysis of clinically acquired CT-scans. We retrospectively included patients in our center that received any CT-scan including the abdomen and had a 24-h urine collection within 2 weeks of the scan (n = 636). To validate the equations in healthy individuals, we included a kidney donor dataset (n = 287). We used a deep learning algorithm to segment muscle and fat at the 3rd lumbar vertebra, calculate surface areas and extract radiomics parameters. Two equations for CT-based estimate of RenAl FuncTion (CRAFT 1 including CT parameters, age, weight, and stature and CRAFT 2 excluding weight and stature) were developed and compared to the Cockcroft-Gault and the Ix equations. CRAFT1 and CRAFT 2 were both unbiased (MPE = 0.18 and 0.16 mmol/day, respectively) and accurate (RMSE = 2.68 and 2.78 mmol/day, respectively) in the patient dataset and were more accurate than the Ix (RMSE = 3.46 mmol/day) and Cockcroft-Gault equation (RMSE = 3.52 mmol/day). In healthy kidney donors, CRAFT 1 and CRAFT 2 remained unbiased (MPE = − 0.71 and − 0.73 mmol/day respectively) and accurate (RMSE = 1.86 and 1.97 mmol/day, respectively). Deep learning-based extraction of body-composition parameters from abdominal CT-scans can be used to reliably estimate creatinine production in both patients as well as healthy individuals. The presented algorithm can improve the estimation of renal function in patients who have recently had a CT scan. The proposed methods provide an improved estimation of renal function that is fully automatic and can be readily implemented in routine clinical practice.

Liver Enhancement on Computed Tomography Is Suboptimal in Patients with Liver Steatosis

This study's aim was twofold. Firstly, to assess liver enhancement quantitatively and qualitatively in steatotic livers compared to non-steatotic livers on portal venous computed tomography (CT). Secondly, to determine the injection volume of contrast medium in patients with severe hepatic steatosis to improve the image quality of the portal venous phase. We retrospectively included patients with non-steatotic (n = 70), the control group, and steatotic livers (n = 35) who underwent multiphase computed tomography between March 2016 and September 2020. Liver enhancement was determined by the difference in attenuation in Hounsfield units (HU) between the pre-contrast and the portal venous phase, using region of interests during in three different segments. Liver steatosis was determined by a mean attenuation of ≤40 HU on unenhanced CT. Adequate enhancement was objectively defined as ≥50 ΔHU and subjectively using a three-point Likert scale. Enhancement of non-steatotic and steatotic livers were compared and associations between enhancement and patient- and scan characteristics were analysed. Enhancement was significantly higher among the control group (mean 51.9 ± standard deviation 11.5 HU) compared to the steatosis group (40.6 ± 8.4 HU p for difference < 0.001). Qualitative analysis indicated less adequate enhancement in the steatosis group: 65.7% of the control group was rated as good vs. 8.6% of the steatosis group. We observed a significant correlation between enhancement, and presence/absence of steatosis and grams of iodine per total body weight (TBW) (p < 0.001; adjusted R2 = 0.303). Deduced from this correlation, theoretical contrast dosing in grams of Iodine (g I) can be calculated: g I = 0.502 × TBW for non-steatotic livers and g I = 0.658 × TBW for steatotic livers. Objective and subjective enhancement during CT portal phase were significantly lower in steatotic livers compared to non-steatotic livers, which may have consequences for detectability and contrast dosing.

Domain-Adaptive Artificial Intelligence-Based Model for Personalized Diagnosis of Trivial Lesions Related to COVID-19 in Chest Computed Tomography Scans

BACKGROUND

Early and accurate detection of COVID-19-related findings (such as well-aerated regions, ground-glass opacity, crazy paving and linear opacities, and consolidation in lung computed tomography (CT) scan) is crucial for preventive measures and treatment. However, the visual assessment of lung CT scans is a time-consuming process particularly in case of trivial lesions and requires medical specialists.

METHOD

A recent breakthrough in deep learning methods has boosted the diagnostic capability of computer-aided diagnosis (CAD) systems and further aided health professionals in making effective diagnostic decisions. In this study, we propose a domain-adaptive CAD framework, namely the dilated aggregation-based lightweight network (DAL-Net), for effective recognition of trivial COVID-19 lesions in CT scans. Our network design achieves a fast execution speed (inference time is 43 ms on a single image) with optimal memory consumption (almost 9 MB). To evaluate the performances of the proposed and state-of-the-art models, we considered two publicly accessible datasets, namely COVID-19-CT-Seg (comprising a total of 3520 images of 20 different patients) and MosMed (including a total of 2049 images of 50 different patients).

RESULTS

Our method exhibits average area under the curve (AUC) up to 98.84%, 98.47%, and 95.51% for COVID-19-CT-Seg, MosMed, and cross-dataset, respectively, and outperforms various state-of-the-art methods.

CONCLUSIONS

These results demonstrate that deep learning-based models are an effective tool for building a robust CAD solution based on CT data in response to present disaster of COVID-19.