Clinical Application of Artificial Intelligence-Based Computed Tomography Analysis of Myosteatosis in Localized Renal Cell Carcinoma

Article information

J Urol Oncol. 2024;22(3):237-245

Publication date (electronic) : 2024 November 30

doi : https://doi.org/10.22465/juo.244800880044

Byeong Jin Kang¹, Kyung Hwan Kim¹, Seung Baek Hong², Nam Kyung Lee², Suk Kim², Sihwan Kim^3,4, Hong Koo Ha^,1

¹Department of Urology, Biomedical Research Institute, Pusan National University Hospital, Pusan National University School of Medicine, Busan, Korea

²Department of Radiology, Biomedical Research Institute, Pusan National University Hospital, Pusan National University School of Medicine, Busan, Korea

³Department of Applied Bioengineering, Graduate School of Convergence Science and Technology, Seoul National University, Seoul, Korea

⁴ClariPi Research, Seoul, Korea

Corresponding author: Hong Koo Ha Department of Urology, Biomedical Research Institute, Pusan National University Hospital, Pusan National University School of Medicine, 179 Gudeok-ro, Seo-gu, Busan 49241, Korea Email: hongkooha@pusan.ac.kr

Received 2024 September 11; Revised 2024 November 6; Accepted 2024 November 19.

Abstract

INTRODUCTION

Renal cell carcinoma (RCC) is one of the most prevalent malignancies affecting kidneys accounting for approximately 90% of all kidney cancers. Although advancements in early detection and surgical techniques have improved the prognosis of patients with localized RCC, identifying additional prognostic factors remains crucial to refine risk stratification and treatment approaches [1].

Recently, the growing focus has been placed on body composition, particularly sarcopenia and myosteatosis, as important predictors of cancer outcomes. Sarcopenia, defined as the loss of skeletal muscle mass and strength, has been extensively studied and is linked to poor oncologic outcomes in various cancers, including pancreatic and gastrointestinal malignancies [2]. Myosteatosis, characterized by increased fat infiltration within the muscle tissue, represents not only a loss of muscle mass but also a decrease in muscle quality, which has been associated with adverse outcomes, such as reduced survival and increased postoperative complications [3].

The prognostic impact of myosteatosis has been well documented in gastric, pancreatic, and metastatic cancers, where the condition has been linked to poor survival and treatment-related toxicity [4,5]. Additionally, studies examining the variability in skeletal muscle radiodensity in patients with metastatic cancer have demonstrated that muscle quality, including radiodensity, is significantly associated with adverse outcomes, further supporting the significance of body composition in cancer prognosis [6]. Despite its significance in other cancers, the role of myosteatosis in RCC remains unclear. Although sarcopenia has been studied in RCC, a notable lack of dedicated research exists examining the specific impact of myosteatosis on the survival and outcomes of patients with RCC. This gap in the literature underscores the need for further investigations into the prognostic significance of myosteatosis in RCC.

Manual and semiautomated methods used to assess myosteatosis have limitations including labor intensity, variability, and lack of accuracy, which can hinder consistent results [7,8]. To overcome these challenges, we utilized an artificial intelligence (AI)-based approach to analyze preoperative computed tomography (CT) scans and evaluate myosteatosis in patients with RCC. Although the use of AI-driven body composition analysis tools is becoming increasingly common in oncology, their application in RCC remains limited. By employing this automated method, we aimed to enhance the precision and reproducibility of the analysis. This study further assessed the prognostic value of AI-based myosteatosis evaluation for predicting disease-free survival (DFS) and adverse pathological outcomes in patients with localized RCC undergoing nephrectomy, addressing a critical gap in the literature.

MATERIALS AND METHODS

This retrospective cohort study aimed to evaluate the prognostic impact of myosteatosis on DFS and pathological outcomes in patients with localized RCC who underwent partial nephrectomy. Preoperative CT scans were analyzed using an AI-based tool to evaluate skeletal muscle quality, with a focus on the presence of myosteatosis. Clinical and pathological data were collected from the patient’s medical records. The primary objective of this study was to determine whether myosteatosis is an independent predictor of oncologic outcomes in this patient cohort.

This study included patients who underwent surgery between 2011 and 2017. Patients diagnosed with localized RCC and treated with a partial nephrectomy were included. To be eligible, patients were required to undergo preoperative abdominal CT scans performed within 3 months before surgery. Noncontrast axial-view CT images were specifically utilized for body composition analysis to assess myosteatosis. Patients were excluded if they had incomplete clinical data or missing follow-up information, which could compromise the accuracy of survival and pathological outcome analyses. Additionally, patients with metastatic RCC at the time of diagnosis were excluded to specifically examine the effect of myosteatosis on the localized RCC cohort.

The demographic data included age, sex, and body mass index (BMI). Furthermore, BMI was classified according to the Korean Society for the Study of Obesity guidelines, which define overweight as a BMI≥23 kg/m² and obesity as a BMI ≥25 kg/m² [9]. These thresholds were based on the increased risk of metabolic complications in the Korean population, even at low BMI. In addition to BMI, comorbidities and medical history, including the presence of hypertension and diabetes, were documented for each patient.

Preoperative abdominal CT scans obtained within 3 months before surgery were analyzed using the AI-based muscle analysis software (ClariSarco, ClariPi, Seoul, Korea; https://claripi.com/clarisarco2). This software used 2 deep learning models to assess the abdominal muscle composition in CT images. When the CT data is input into the tool, the first model localizes the vertebral bodies (T12–L4) on the reconstructed image by maximum intensity projection method. Once the L3 level is identified, a second model automatically segments the abdominal muscle areas with a 2-dimensional U-Net architecture (Fig. 1).

Fig. 1.

Schematics for abdomen muscle analysis algorithm in ClariSarco. The abdominal computed tomography (CT) image is used for both finding the location of third lumbar spine (L3) level and segmentation of the abdominal muscular regions. For the L3-level localization (green rectangular box), the CT image is projected to anterior-posterior direction using maximum intensity projection (MIP) method, and the MIP image is input to the convolutional neural network (CNN)-based detection module. Simultaneously, the CT image is segmented by UNet-based CNN-module generating abdominal muscular region mask. Intermuscular adipose tissue area (IMATA), low attenuation muscle area (LAMA) and normal attenuation area (NAMA) were defined within the muxcular region based on CT-density (Hounsfield unit). All the information was used for calculating the skeletal muscle index and myosteatosis percentage defined as (LAMA+NAMA)/height2 and (1-NAMA/TAMA) ×100, respectively. AI, artificial intelligence; SMA, skeletal muscle area.

The segmentation process assesses key muscle components, including intermuscular adipose tissue (IMAT), low attenuation muscle area (LAMA), normal attenuation muscle area (NAMA), skeletal muscle area (SMA), and total abdominal muscle area (TAMA). These areas are defined by Hounsfield unit (HU) ranges: NAMA (30 to 150 HU), LAMA (-29 to 29 HU), and IMAT (-190 to -30 HU). The SMA is the sum of the LAMA and NAMA, whereas the TAMA includes the IMAT, LAMA, and NAMA. Utilizing these measurements, BMI, skeletal muscle index, and myosteatosis percentage (calculated as 100×[1-NAMA/TAMA]) were derived for each patient (Fig. 2) [10,11].

Fig. 2.

Abdominal muscle segmentation and tissue attenuation analysis. This figure illustrates abdominal muscle segmentation at the L3 vertebral level using computed tomography (CT) imaging. The tissue attenuation histogram shows the distribution of muscle density, with the areas categorized into intermuscular adipose tissue (IMAT), low attenuation muscle area (LAMA), and normal attenuation muscle area (NAMA). HU, Hounsfield unit; BMI, body mass index; SMI, skeletal muscle index.

Myosteatosis was assessed using the NAMA-to-TAMA ratio with predefined sex-specific cutoff values derived from the study of Kim et al. [12] on a Korean population. Based on their findings, we used a threshold of 33.6% or higher for male patients and 34.9% for female patients to reflect significant fat infiltration within muscle tissue. These cutoff values were established based on a T score of-2, reflecting a clinically relevant level of fat infiltration. This differentiation considers the physiological differences in muscle composition between sexes. A low NAMA-to-TAMA ratio indicates a high degree of fat infiltration, signifying poor muscle quality and an increased likelihood of myosteatosis (Fig. 3).

Fig. 3.

Representative computed tomography (CT) analysis of myosteatosis and no myosteatosis groups. Comparison of myosteatosis (bottom) and no myosteatosis (top) groups. The myosteatosis case shows a reduced normal attenuation muscle area (NAMA) proportion within the total abdominal muscle area (TAMA), indicating greater fat infiltration in muscle tissue. SMA, skeletal muscle area; HU, Hounsfield unit; LAMA, low attenuation muscle area; IMAT, intermuscular adipose tissue.

Patients were divided into 2 groups based on the presence or absence of myosteatosis to compare pathologic results and prognostic outcomes. Key characteristics, such as tumor site, histology, tumor size, Fuhrman/World Health Organization grade, and extent of invasion (including capsule invasion, extension into the perirenal fat, and lymphovascular invasion) were recorded. The primary tumor (T) stage was determined according to the TNM classification. Additionally, DFS was compared between the myosteatosis and no myosteatosis groups to assess the impact of the condition on patient prognosis.

Descriptive statistics were utilized to summarize patient demographics and tumor characteristics, with continuous variables presented as means and standard deviations or medians and interquartile ranges, and categorical variables as frequencies and percentages. The Kaplan-Meier method was used to analyze DFS, with comparisons between the myosteatosis and no myosteatosis groups using the log rank test. Cox proportional hazard regression was employed to identify independent predictors of DFS, adjusting for potential confounders such as age, BMI, and tumor stage. Statistical significance was set at p<0.05.

RESULTS

1. Preoperative Patient Characteristics

In this study of 170 patients, 36 (21.2%) were identified with myosteatosis, while 134 (78.8%) did not have the condition (Fig. 3). A significant difference was observed in sex distribution, with a higher proportion of females in the myosteatosis group (61.1%) compared to males (38.9%). In contrast, the no myosteatosis group consisted predominantly of males (76.9%). Patients with myosteatosis were significantly old, with an average age of 64.2 years versus 57.5 years in those without myosteatosis (p<0.05) (Table 1).

Table 1.

Preoperative patients’ characteristics

BMI also differed between the groups, with a higher percentage of patients with myosteatosis classified as overweight or obese (83.3%) than those without myosteatosis (59.7%) (p<0.05). A chi-square test of the BMI distribution across the 3 categories (normal, overweight, and obese) showed a statistically significant difference between the myosteatosis and no myosteatosis groups (p<0.05), indicating that the overall BMI distribution was associated with the presence of myosteatosis. Additionally, the correlation analysis between BMI and myosteatosis percentage showed a statistically significant positive association (Pearson correlation coefficient r=0.319, p<0.001), indicating that patients with a higher BMI tended to exhibit higher levels of myosteatosis. The proportion of patients with an elevated neutrophil lymphocyte ratio (≥3.4) was higher in the myosteatosis group than in the no myosteatosis group; however, this difference was not statistically significant. Overall, myosteatosis was associated with old age, female sex, and an elevated BMI in this patient cohort (Table 1).

2. Pathologic Outcomes

Pathological characteristics of the myosteatosis and no myosteatosis groups were largely comparable. No significant differences were observed in the pT stage distribution (p=0.286) or histologic type (p=0.526), with clear cell carcinoma being the most common type in both groups. The tumor grade also exhibited no significant difference, with similar proportions of low (G1 and G2) and high-grade (G3 and G4) tumors between the groups (p=0.824) (Table 2).

Table 2.

Comparison of pathological characteristics between patients with and without myosteatosis after surgery

However, lymphovascular invasion was significantly more frequent in the myosteatosis group (22.2%) than in the no myosteatosis group (9%) (p<0.05). Furthermore, tumor necrosis was also prevalent in the myosteatosis group (27.8% vs. 5.2%, p<0.05). Other measures of tumor invasion, including renal capsule and perirenal fat invasion, did not differ significantly between the groups (Table 2). In the ccRCC subgroup analysis, the myosteatosis group also demonstrated higher rates of lymphovascular invasion (26.7% vs. 7.8%, p<0.05) and tumor necrosis (30% vs. 5.9%, p<0.05) compared to the no myosteatosis group (Table 3)

Table 3.

Comparison of pathological characteristics between clear cell renal cell carcinoma patients with and without myosteatosis after surgery

3. Survival Analysis

In the analysis of DFS among 170 patients, those with myosteatosis experienced shorter DFS than those without myosteatosis, with median follow-up times of 98.5 months and 101.5 months, respectively. Recurrences were frequent and occurred earlier in the myosteatosis group. Kaplan-Meier survival analysis and log-rank test demonstrated a statistically significant difference in DFS between the 2 groups (p<0.05) (Fig. 4).

Fig. 4.

Kaplan-Meier analysis of disease-free survival (DFS) in patients with and without myosteatosis. Kaplan-Meier curves showing DFS for patients with myosteatosis (red line) and those without myosteatosis (blue line). A statistically significant difference in DFS was observed between the 2 groups (log-rank p<0.05).

In univariate analysis, age, myosteatosis, and margin positivity were significantly associated with DFS (p<0.05). In multivariate Cox regression analysis, myosteatosis remained an independent predictor of DFS with a hazard ratio (HR) of 1.50 (95% confidence interval [CI], 1.18–1.91, p=0.040) [3]. Age was not statistically significant in the multivariate model (p=0.362). Margin positivity was also independently associated with poor DFS (HR, 3.25; 95% CI, 1.14–8.12; p<0.001). Additionally, sinus fat invasion displayed a marginally significant association with DFS (HR, 1.63; 95% CI, 0.57–4.57; p=0.024) (Table 4).

Table 4.

Univariate and multivariate analysis of disease-free survival in patients with renal cell carcinoma

DISCUSSION

This study highlights that myosteatosis evaluated using AI-driven CT analysis is associated with higher rates of adverse pathological features, such as lymphovascular invasion and tumor necrosis, and is an independent predictor of DFS in patients with localized RCC. The use of AI for preoperative body composition assessment offers a precise and reproducible method for identifying high-risk patients, contributing valuable insights into the limited research on myosteatosis in patients with RCC.

In this study, traditional pathological features, such as tumor necrosis, positive surgical margin, and lymphovascular invasion, which are typically associated with poor oncologic outcomes, showed limited influence on DFS in our early stage RCC cohort treated with partial nephrectomy. Previous studies, including those by Chang et al. [13] and Mouracade et al. [14], have similarly found that these adverse pathological features may have a reduced prognostic impact in early stage, curatively treated RCC.

Previous research has demonstrated the negative impact of myosteatosis on cancer outcomes in various malignancies, including pancreatic and colorectal cancers. For instance, Kim et al. [2] revealed that both sarcopenia and myosteatosis are significant predictors of poor prognosis in patients with resectable pancreatic ductal adenocarcinoma. Similarly, Sueda et al. [15] discovered that low muscularity and myosteatosis were associated with poor long-term outcomes in patients undergoing curative surgery for colorectal cancer, further emphasizing the detrimental effects of muscle quality on oncological outcomes. The findings of our study are consistent with those of previous studies, in which myosteatosis was consistently linked to poor outcomes. However, this study was one of the first to focus on the prognostic significance of myosteatosis, specifically in localized RCC, highlighting the need for further studies in this area.

In our cohort, a relatively higher prevalence of myosteatosis was observed among female patients. This finding aligns with previous studies that highlight a stronger correlation between obesity and RCC risk in women compared to men. For instance, the Me-Can study indicated that high BMI significantly increased RCC risk in women more than in men, suggesting that metabolic factors related to obesity may contribute to this sex-specific difference in myosteatosis prevalence [16].

Multiple interrelated mechanisms support the association between myosteatosis and poor cancer outcomes. Myosteatosis is characterized by fat infiltration into skeletal muscles and leads to metabolic dysregulation, including insulin resistance, increased systemic inflammation, and oxidative stress, as noted by Miljkovic and Zmuda [17] and Miljkovic et al. [18]. This environment promotes cancer progression and weakens the immune response, thereby contributing to cancer cachexia and worsening survival outcomes. Additionally, Sueda et al. [15] emphasized that the skeletal muscle acts as an endocrine organ, releasing cytokines that maintain homeostasis. However, tumor-driven proinflammatory cytokines such as interleukin (IL)-1β, IL-6, and tumor necrosis factor-alpha disrupt this balance, leading to muscle degradation and exacerbating cancer-related muscle loss. Tan et al. [19] observed that in localized RCC, higher visceral adiposity correlated with larger tumor sizes, suggesting a potential link between body composition and disease aggression. Similarly, Rysz et al. [20] proposed that metabolic factors like insulin resistance may foster a tumor promoting environment through chronic inflammation, contributing to adverse pathological features such as tumor necrosis and lymphovascular invasion. Collectively, these mechanisms demonstrate that systemic inflammation and metabolic disturbances caused by myosteatosis are key drivers of poor oncologic outcomes in patients with cancer, further underscoring the need for preoperative assessment of body composition to improve the prognosis and treatment strategies for RCC.

This study highlights the potential of preoperative CT-based AI analysis to improve risk stratification in patients with RCC. Although previous research has largely focused on BMI, sarcopenia, and visceral fat [21], few studies have explored muscle quality using AI-driven CT analysis. By assessing myosteatosis, which is a predictor of poor outcomes, we offer a precise tool for identifying high-risk patients who may benefit from aggressive treatment or frequent follow-up.

Recent studies have demonstrated the usefulness of AI-based body composition analysis in improving cancer prognosis. For instance, Borrelli et al. [22] employed AI-powered CT software to standardize body composition assessments in patients with urothelial tumors, correlating sarcopenia with patient outcomes. Similarly, Xu et al. [23] demonstrated that AI algorithms applied to CT scans enhanced the risk prediction for lung cancer mortality and all-cause mortality by incorporating muscle quality metrics such as skeletal muscle attenuation. These studies underscore the growing importance of AI-driven assessments in the refinement of prognostic models. Our study applied this approach to RCC, demonstrating the utility of AI-driven analysis of muscle quality in improving prognosis alongside traditional clinical factors. By identifying patients with myosteatosis, treatment plans can be refined, and monitoring enhanced. Consistent with the findings of Westenberg et al. [8], our use of AI-based analysis provided improved accuracy, efficiency, and reproducibility over manual methods, thereby reducing observer bias and saving time.

Although our study did not focus on lifestyle interventions, findings from the ReLife study by Maurits et al. [24] suggest that managing lifestyle factors including diet and physical activity may improve outcomes for patients with poor muscle quality. The AI-driven muscle quality assessment used in our study could help identify patients who might benefit from such strategies, offering a pathway for improved oncologic outcomes through personalized care.

The retrospective design of this study and the use of single center data may have introduced a selection bias, limiting the generalizability of the results. Additionally, the follow up period may not have fully captured the long-term impact of myosteatosis in patients with RCC. To address these limitations, future multicenter prospective studies with prolonged follow-up periods and diverse patient populations are needed to validate our findings and further explore the biological mechanisms linking myosteatosis to poor outcomes, particularly in terms of metabolic dysregulation and fat infiltration in the muscle tissue.

CONCLUSIONS

This study demonstrated that myosteatosis, assessed using AI-driven CT analysis, is associated with adverse pathological features in localized RCC, including lymphovascular invasion and tumor necrosis. Additionally, myosteatosis was independently associated with reduced DFS. AI-based body composition analysis of myosteatosis is a reliable method for improving risk stratification and identifying high-risk patients. Future multicenter studies with extended follow-up periods are required to validate these findings and further explore the underlying mechanisms.

Article information Continued