Skip to main content
NCBI home page
Journal of the Korean Society of Radiology logoLink to Journal of the Korean Society of Radiology
. 2025 Jul 25;86(4):501–511. doi: 10.3348/jksr.2023.0108

Enhancing the Diagnosis of Sarcopenia Through Low-Dose Chest CT and Artificial Intelligence-Based Segmentation: Optimizing Resource Utilization in Healthcare

저선량 흉부 CT와 인공지능 분석을 통한 근감소증 진단의 향상: 보건의료 자원 이용의 최적화

Hyeon Tak Lee 1,†, Jung Keun Hyun 2,†, Sang Yoon Kim 1,✉
PMCID: PMC12328913  PMID: 40787303

Abstract

Purpose

We aimed to apply artificial intelligence (AI)-based segmentation techniques to low-dose chest CT images to measure the muscle mass and investigate whether these results can be helpful in diagnosing sarcopenia.

Materials and Methods

We retrospectively included 100 participants who had undergone both routine-dose contrast-enhanced chest CT (routine CT) and low-dose chest CT (LDCT) within 6 months of the study. Muscle segmentation and measurement were performed using a commercially available AI-based software. Skeletal muscle volume index (SMVI, g/m2) and skeletal muscle index (L1 SMI, cm2/m2) were measured on CT images using the software. We compared the SMVI obtained through routine CT and LDCT in the same patients and investigated the correlation between SMVI and L1 SMI.

Results

The SMVI obtained through both routine CT and LDCT demonstrated statistically significant associations with the conventional L1 SMI (r = 0.89, 0.85, p < 0.001). The SMVI obtained through LDCT showed a slightly lower value but maintained a high correlation with SMVI obtained through routine CT (r = 0.956, p < 0.001).

Conclusion

Utilization of the LDCT protocol in diagnosing sarcopenia appears to be a valuable approach. By applying AI-based segmentation to these scans, it has become possible to accurately measure the entire muscle mass.

Keywords: Sarcopenia, Body Composition, Computed Tomography, Artificial Intelligence

Graphical Abstract

graphic file with name jksr-86-501-abf001.jpg

INTRODUCTION

Sarcopenia, defined as age-related loss of muscle mass (1), is a known prognostic factor associated with poor survival (2) and patients’ quality of life (3). Despite its significant clinical and public health burden, there are currently no standardized diagnostic criteria for sarcopenia, thus making its evaluation challenging. Various methods are employed for evaluate muscle mass, including anthropometry, bioelectrical impedance analysis (BIA), and medical imaging. While anthropometry and BIA are convenient and non-invasive methods for estimating body composition, they have limitations, such as potential inaccuracy due to measurement errors. Dual-energy X-ray absorptiometry (DEXA) is currently the most widely used medical imaging test for relatively accurately measurement of muscle mass. It offers an advantage by enabling the measurement of body fat and bone density. However, it is challenging to differentiate trunk muscles from the surrounding fat and viscera; hence, the diagnosis of sarcopenia using a DEXA scan primarily relies on measuring the limb muscle mass. Although limb muscle mass is a valuable indicator of muscle health and can provide insights into the overall changes in muscle mass, relying solely on limb muscle mass measurements may not depict the full extent of muscle loss in individuals with sarcopenia.

CT has emerged as a promising tool for assessing sarcopenia due to its ability to accurately quantify muscle mass and density. CT can accurately measure the muscle mass of the trunk compared to DEXA. Generally, the skeletal muscle area (SMA) and height-adjusted SMA (skeletal muscle index [SMI]), measured on an axial CT image at the third lumbar vertebral (L3) level, are known to aid in predicting the diagnosis of sarcopenia and its associated morbidity and mortality in malignant and non-malignant diseases (4,5). Furthermore, research studies are actively being conducted to measure not only the muscle area in a single section but also the muscle volume of the entire scanning range using artificial intelligence (AI)-based segmentation for sarcopenia measurement (6,7,8,9).

However, CT scans expose the patients to more radiation compared to DEXA, making them less suitable for proactive diagnosis of sarcopenia. Most studies on the diagnosis of sarcopenia using CT images have been retrospective analyses of images obtained for diagnoses of other diseases. Therefore, minimizing the radiation dose of CT scans is an important consideration in researching the diagnosis of sarcopenia. Currently, there is inadequate research discussing the use of CT with a low-dose protocol for evaluating sarcopenia.

This study focuses on utilizing an AI segmentation program to measure the total muscle volume on chest CT scans instead of relying on a single cross-sectional muscle area. It also includes a comparison of the muscle volume between low-dose and routine-dose chest CT scans in the same patients and aims to investigate the correlation between the muscle volume obtained through volume segmentation and SMI.

MATERIALS AND METHODS

PARTICIPANTS AND CT TECHNIQUES

The present study was approved by the Institutional Review Board (IRB No. 2022-12-022-002). From July 2022 to December 2022, we retrospectively included consecutive participants who had undergone both routine CT and low-dose chest CT (LDCT) scans for any medical purpose at a single tertiary institution. We excluded patients with conditions likely to induce changes in muscle mass (such as cancer or pneumonia) during the research period. The study primarily focused on patients undergoing examinations for the follow-up of small pulmonary nodules. In our institute, routine and low-dose chest CT scans generally cover the scan range from the C6–7 to the L1–2 vertebral levels. For consistency in the results, this study analyzed muscle mass from the T1 to the L1 vertebral levels. Routine CT scans were performed using a 256-channel CT scanner (Ict, Philips Healthcare, Amsterdam, The Netherlands) with the following parameters: tube voltage, 120 kVp; effective tube current, 88–159 mAs; slice thickness, 2.5 mm; and pitch, 0.6–0.859. Intravenous administration of a contrast agent was performed at a rate of 3–4 Ml/s. LDCT was performed with the same 256-channel CT scanner (Ict, Philips Healthcare) with iterative reconstruction (IMR1 Sharplus) with the following parameters: tube voltage, 120 kVp; effective tube current, 28 mAs; slice thickness, 2.5 mm; and pitch, 0.758–0.797. This study used a CT image matrix of 512 × 512, with a field of view ranging from 290 to 350 mm. The estimated pixel size was between 0.56 and 0.68 mm. All scans were performed in a supine position.

IMAGE PROCESSING AND DATA ANALYSIS

Muscle segmentation was performed using a commercially available AI-based software (DeepCatch v1.1.7, Medical IP CO. LTD., Seoul, South Korea). It is a deep neural network-based software for automatic volumetric segmentation of body composition from anonymized CT images in DICOM format. The software is based on a neural network which is trained using 39268 image slices, providing an average dice score of 96.8%–99.2% for muscles in the domestic validation set, showing high dice similarity coefficient (DSC) results (9). Two radiologists (one dedicated musculoskeletal radiologist and one third-year radiology resident) reviewed the CT images. An axial image was selected at the L1 level for SMA measurement. At the L1 level, the measurement included the erector spinae, latissimus dorsi, external and internal oblique muscles, rectus abdominis, and intercostal muscles. For the measurement of the entire muscle volume on the CT scan, the segmentation range was set between the level of the upper endplate of T1 vertebra and the lower endplate of L1 vertebra (Fig. 1).

Fig. 1. Muscle segmentation using chest CT images and artificial intelligence-based software.

Fig. 1

Open in a new tab

A-F. Low-dose chest CT slice with iterative reconstruction at the L1 level is selected (A) and automated muscle segmentation of the slice is performed (B). A coronal reconstructed image shows the range of axial image selection (C). The same patient’s routine dose contrast-enhanced chest CT slice at the L1 level is selected (D) and automated muscle segmentation of the slice is performed (E). A 3D-reconstructed image of the selected muscle area is visualized (F).

The software program computed SMA (cm2) by summing the cross-sectional muscle areas and multiplying by the pixel surface area. Skeletal muscle index (L1 SMI) was calculated by dividing SMA by the height squared (cm2/m2). Skeletal muscle volume (SMV, g), obtained from volume segmentation, was normalized to height squared (m2), and it was reported as SMVI (g/m2).

We use the paired t-test to compare the SMVI obtained using LDCT with that obtained using routine CT scans in the same patients. Subsequently, Pearson’s correlation coefficients and linear regression analysis were applied to assess the correlation between these parameters. Additionally, we explored the relationship between the SMVI and SMI using Pearson’s correlation coefficients and linear regression analysis, setting the statistical significance at p < 0.05 (SPSS v12.0, Chicago, IL, USA).

RESULTS

Study population

Among the 134 participants who underwent both routine CT and LDCT scans, 34 were excluded due to the following reasons: scanning with different CT scanners (n = 15), incomplete patient data of height and weight (n = 10), gross difference in the muscle volume on time interval perceived by reviewers (n = 9). Finally, CT scans from 100 participants (mean age, 64 years; standard deviation [SD] ± 14 years; 66 males and 34 females) were included in the analysis. Mean height and weight of the participants were 163.5 ± 8.4 cm and 64.5 ± 15.9 kg, respectively.

Comparison of L1 SMI and SMVI

The correlation between L1 SMI and SMVI obtained through both routine CT and LDCT was statistically significant (p < 0.001), and both values were high (r = 0.89, 0.85). The correlation coefficient for SMVI obtained through routine CT and L1 SMI was slightly stronger (r = 0.89) than that for SMVI obtained through LDCT and L1 SMI (r = 0.85). The linear correlation is shown in Fig. 2; L1 SMI = 3.636 + 0.025 × SMVI (routine CT, A), L1 SMI = 5.19 + 0.026 × SMVI (LDCT, B).

Fig. 2. The scatter plot of L1 SMI and SMVI is obtained through routine CT (A); L1 SMI and SMVI are obtained through LDCT (B).

Fig. 2

Open in a new tab

A. A significant linear correlation is observed between L1 SMI and SMVI obtained through routine CT (r = 0.89, p < 0.001).

B. A significant linear correlation is also observed between L1 SMI and SMVI obtained through LDCT (r = 0.85, p < 0.001).

LDCT = low-dose chest CT, SMI = skeletal muscle index, SMVI = skeletal muscle volume index

Comparison of SMVI obtained through LDCT and routine CT scans

SMVI obtained through routine CT (mean = 1476.7, SD = 354.6) was higher than that obtained through LDCT (mean = 1329.2, SD = 318.5; p < 0.001). A correlation between the SMVI obtained through routine CT and LDCT was also observed (r = 0.956, p < 0.001). Linear correlation is shown in Fig. 3; SMVI (routine CT) = 76.37 + 1.05 × SMVI (LDCT).

Fig. 3. The scatter plot of SMVI is obtained through routine CT and SMVI obtained through LDCT.

Fig. 3

Open in a new tab

A significant linear correlation between the SMVI values obtained through routine CT and LDCT is confirmed (r = 0.956, p < 0.001).

LDCT = low-dose chest CT, SMVI = skeletal muscle volume index

DISCUSSION

Accurate measurement of the muscle mass is an essential element for the diagnosis of sarcopenia. Although the Asian Working Group for Sarcopenia (AWGS) has suggested more cost-effective choices, such as DEXA or BIA, instead of CT, it is important to note that CT scans have a distinct ability to differentiate between the muscle and surrounding tissues. This feature provides a strong advantage for measuring the muscle mass in areas like the torso, where various tissues are intermixed. However, CT scans are not actively utilized for the diagnosis of sarcopenia due to their higher cost compared to DEXA and the associated higher radiation exposure. Currently, studies on the diagnosis of sarcopenia using CT imaging have mostly been retrospective analyses of CT scans performed for diagnosing other conditions.

In terms of methodology, the diagnosis of sarcopenia through CT based on the AWGS and the European Working Group on Sarcopenia in Older People guidelines is based on the quantification of the cross-sectional area of skeletal muscle at the L3 level (SMI) (10,11). However, it should be noted that this method relies solely on predicting the total muscle mass based on a single cross-sectional muscle area at the L3 level, which has its own limitation. It is believed that measuring the volume of muscles within the scanning range, rather than relying on a single cross-sectional area, would more accurately predict the overall muscle mass. Consequently, research efforts have been recently made to measure the total muscle mass across all images included in a CT scan and utilize this information. Furthermore, some studies have incorporated AI technology aiming for accuracy and timesaving (8,12). In this study, we applied a commercially available AI-based software (DeepCatch v1.1.7, Medical IP) to analyze chest CT scans. The DICOM files were analyzed using the equipment provided by the company, and although the analysis time was longer than that for measuring only the SMA in a single cross-section, we were able to measure the entire muscle volume included in the CT scan in a relatively short time of approximately 4 to 5 minutes. This software can provide analysis information on the other aspects of body composition, such as skeleton and adipose tissue, in addition to information on the muscle mass. In our study, while determining the SMI, we chose to calculate the SMA at the L1 level instead of the L3 level, as the L3 level was not included in the chest CT protocol. Traditionally, various studies have utilized the L3 SMI obtained through abdominal CT for diagnosing sarcopenia. However, recent research has explored alternative SMI values obtained at different levels, including L1, as viable options (13,14,15).

While considering the radiation dose, a lower radiation dose of CT may lead to increased image noise, resulting in reduced spatial resolution. This can pose challenges for AI segmentation programs to accurately identify and delineate the muscle boundaries, potentially leading to underestimation of the muscle volume. In this study, Iterative Model Reconstruction technology was used to reconstruct the images, with the aim to improve low-contrast detectability and image noise (16,17). Despite using this technique, the muscle volume was somewhat underestimated. Consequently, in situations where iterative reconstruction is not available, an increase in the radiation dose might be necessary for accurately evaluating the muscle volume. When comparing the calculated SMVI between routine CT (mean = 1476.7, SD = 354.6) and LDCT (mean = 1329.2, SD = 318.5) using a paired t-test, we observed that the SMVI obtained through routine CT was statistically higher by approximately 10%. However, a strong correlation was found between SMVI obtained through routine CT and LDCT (r = 0.956, p < 0.001). Additionally, it is worth noting that the SMVI obtained through either routine CT (r = 0.89) or LDCT (r = 0.85) demonstrated a significant correlation with the L1 SMI. This result indicates that it is possible to measure the entire muscle mass relatively accurately by analyzing low-dose CT images with minimal radiation exposure and cost burden. This suggests that LDCT could be a practical alternative for diagnosing sarcopenia. In Korea, a considerable number of individuals undergo LDCT scans as part of their health check-up and lung cancer screening. Our study suggests that these LDCT scans, which have already been conducted for lung cancer detection, can serve an additional purpose of evaluation of sarcopenia. This usability of LDCT scans not only provides essential lung health information but also offers a cost-effective opportunity to assess the muscle volume and diagnose sarcopenia in the same session. Furthermore, our study indicates that while developing a CT scan protocol for diagnosis of sarcopenia, it is possible to reduce the radiation dose to the level of the LDCT protocol.

This study has the following limitations: 1) Although we analyzed two similar protocols for the same patients, the measurements may have differed due to the inherent differences between routine CT and LDCT. Specifically, our analysis involved LDCT with non-enhanced images, while routine CT utilized contrast media. The absence of contrast media in LDCT might have resulted in reduced image contrast between muscles and surrounding tissues. Hence, future studies should consider standardizing the use of contrast media. 2) Despite our efforts to minimize the time interval between the two tests (LDCT and routine CT), i.e. up to a maximum of 6 months, various factors could have influenced the changes in patients’ muscle volume during this period. 3) This study focused solely on image analysis and excluded the clinical data. Future research should incorporate different clinical information essential for diagnosing sarcopenia.

In conclusion, utilization of the LDCT protocol in diagnosing sarcopenia appears to be a valuable approach. By applying AI-based segmentation to these scans, it has become possible to accurately measure the entire muscle mass. This not only enables efficient diagnosis but also helps minimize radiation exposure and cost burden associated with the diagnosis of sarcopenia. Embracing these advancements can significantly improve patient care and resource utilization in the medical field.

Footnotes

Author Contributions:
  • Conceptualization, all authors.
  • data curation, all authors.
  • formal analysis, all authors.
  • investigation, all authors.
  • methodology, all authors.
  • project administration, all authors.
  • resources, all authors.
  • software, all authors.
  • supervision, all authors.
  • validation, all authors.
  • writing—original draft, all authors.
  • writing—review & editing, all authors.

Conflicts of Interest: The authors have no potential conflicts of interest to disclose.

Funding: None

References

  • 1.Lee K, Shin Y, Huh J, Sung YS, Lee IS, Yoon KH, et al. Recent issues on body composition imaging for sarcopenia evaluation. Korean J Radiol. 2019;20:205–217. doi: 10.3348/kjr.2018.0479. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Shachar SS, Williams GR, Muss HB, Nishijima TF. Prognostic value of sarcopenia in adults with solid tumours: a meta-analysis and systematic review. Eur J Cancer. 2016;57:58–67. doi: 10.1016/j.ejca.2015.12.030. [DOI] [PubMed] [Google Scholar]
  • 3.Beaudart C, Zaaria M, Pasleau F, Reginster JY, Bruyère O. Health outcomes of sarcopenia: a systematic review and meta-analysis. PLoS One. 2017;12:e0169548. doi: 10.1371/journal.pone.0169548. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Prado CM, Lieffers JR, McCargar LJ, Reiman T, Sawyer MB, Martin L, et al. Prevalence and clinical implications of sarcopenic obesity in patients with solid tumours of the respiratory and gastrointestinal tracts: a population-based study. Lancet Oncol. 2008;9:629–635. doi: 10.1016/S1470-2045(08)70153-0. [DOI] [PubMed] [Google Scholar]
  • 5.Martin L, Birdsell L, Macdonald N, Reiman T, Clandinin MT, McCargar LJ, et al. Cancer cachexia in the age of obesity: skeletal muscle depletion is a powerful prognostic factor, independent of body mass index. J Clin Oncol. 2013;31:1539–1547. doi: 10.1200/JCO.2012.45.2722. [DOI] [PubMed] [Google Scholar]
  • 6.van der Werf A, Langius JAE, de van der Schueren MAE, Nurmohamed SA, van der Pant KAMI, Blauwhoff-Buskermolen S, et al. Percentiles for skeletal muscle index, area and radiation attenuation based on computed tomography imaging in a healthy Caucasian population. Eur J Clin Nutr. 2018;72:288–296. doi: 10.1038/s41430-017-0034-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Derstine BA, Holcombe SA, Ross BE, Wang NC, Su GL, Wang SC. Optimal body size adjustment of L3 CT skeletal muscle area for sarcopenia assessment. Sci Rep. 2021;11:279. doi: 10.1038/s41598-020-79471-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Bedrikovetski S, Seow W, Kroon HM, Traeger L, Moore JW, Sammour T. Artificial intelligence for body composition and sarcopenia evaluation on computed tomography: a systematic review and meta-analysis. Eur J Radiol. 2022;149:110218. doi: 10.1016/j.ejrad.2022.110218. [DOI] [PubMed] [Google Scholar]
  • 9.Lee YS, Hong N, Witanto JN, Choi YR, Park J, Decazes P, et al. Deep neural network for automatic volumetric segmentation of whole-body CT images for body composition assessment. Clin Nutr. 2021;40:5038–5046. doi: 10.1016/j.clnu.2021.06.025. [DOI] [PubMed] [Google Scholar]
  • 10.Chen LK, Woo J, Assantachai P, Auyeung TW, Chou MY, Iijima K, et al. Asian working group for sarcopenia: 2019 consensus update on sarcopenia diagnosis and treatment. J Am Med Dir Assoc. 2020;21:300–307.e2. doi: 10.1016/j.jamda.2019.12.012. [DOI] [PubMed] [Google Scholar]
  • 11.Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T, et al. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing. 2019;48:16–31. doi: 10.1093/ageing/afy169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Gu S, Wang L, Han R, Liu X, Wang Y, Chen T, et al. Detection of sarcopenia using deep learning-based artificial intelligence body part measure system (AIBMS) Front Physiol. 2023;14:1092352. doi: 10.3389/fphys.2023.1092352. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Kottlors J, Zopfs D, Fervers P, Bremm J, Abdullayev N, Maintz D, et al. Body composition on low dose chest CT is a significant predictor of poor clinical outcome in COVID-19 disease - A multicenter feasibility study. Eur J Radiol. 2020;132:109274. doi: 10.1016/j.ejrad.2020.109274. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Derstine BA, Holcombe SA, Ross BE, Wang NC, Su GL, Wang SC. Skeletal muscle cutoff values for sarcopenia diagnosis using T10 to L5 measurements in a healthy US population. Sci Rep. 2018;8:11369. doi: 10.1038/s41598-018-29825-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Kim EY, Kim YS, Park I, Ahn HK, Cho EK, Jeong YM, et al. Evaluation of sarcopenia in small-cell lung cancer patients by routine chest CT. Support Care Cancer. 2016;24:4721–4726. doi: 10.1007/s00520-016-3321-0. [DOI] [PubMed] [Google Scholar]
  • 16.den Harder AM, Willemink MJ, Budde RP, Schilham AM, Leiner T, de Jong PA. Hybrid and model-based iterative reconstruction techniques for pediatric CT. AJR Am J Roentgenol. 2015;204:645–653. doi: 10.2214/AJR.14.12590. [DOI] [PubMed] [Google Scholar]
  • 17.Moon SW, Lee SH, Woo A, Leem AY, Lee SH, Chung KS, et al. Reference values of skeletal muscle area for diagnosis of sarcopenia using chest computed tomography in Asian general population. J Cachexia Sarcopenia Muscle. 2022;13:955–965. doi: 10.1002/jcsm.12946. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Journal of the Korean Society of Radiology are provided here courtesy of Korean Society of Radiology

RESOURCES