Predicting preoperative muscle invasion status for bladder cancer using computed tomography-based radiomics nomogram

One of the most prevalent urological malignancies and the tenth most common malignancy worldwide is bladder cancer (BCa) [1]. The most prevalent histological type of BCa is urothelial carcinoma [2]. Based on the degree of tumor infiltration into the bladder wall, BCa can be classified pathologically as either muscle invasive bladder cancer (MIBC) or non-MIBC (NMIBC) [2, 3]. Transurethral cystectomy for bladder tumor (TURBT) is the recommended course of treatment for NMIBC, whereas MIBC typically necessitates additional therapies such radical cystectomy, adjuvant chemotherapy, and radiotherapy [3]. This suggests that one of the most significant criteria determining therapy choices for the clinical management of BCa [4, 5] is histological muscle invasion.

Currently, cystoscopic biopsy is the main approach for tumor diagnosis and clinical staging [4, 6]. However, accurate preoperative diagnosis of muscle invasiveness is not an easy task. Since biopsies are operator-dependent, incorrect staging may occur due to inadequate biopsy samples or problems with sample quality [6, 7]. According to previous studies, 20–80% of BCa patients are incorrectly staged due to differences in biopsy [8, 9].

The most frequent noninvasive evaluation technique in BCa patients is computed tomography (CT), which can be used to locate the tumor and assess the tumor’s quantity, dimension, relationship to surrounding tissues, and metastasis [9]. However, traditional CT scans cannot be used to evaluate muscle invasion in BCa because they have poor soft-tissue resolution and cannot distinguish between the different layers of the bladder wall. Magnetic resonance (MR) is more accurate than CT, but the high cost, prolonged scanning time, and numerous contraindications of MR prevent it from being used widely [9, 10]. In addition, imaging remains a subjective assessment process based on the radiologist’s experience. Therefore, there is a need to develop more accurate techniques to assess BCa invasiveness.

Radiomics is a new imaging analytical tool that quantifies medical image data to provide information on the morphological features, size, and location of lesions, allowing physicians to make more precise diagnoses, particularly in oncology [5‐7]. Computers extract potentially important information from large amounts of medical image data and use different algorithms to build models to explain the association between images and diseases [11]. These models can help physicians detect tumor lesions and assess the aggressiveness of the lesions, and estimate the prognosis and response to treatment in cancer patients [11, 12]. Previous studies have demonstrated the high diagnostic performance of multi-sequence MR-based radiomics in predicting muscle invasion in BCa [5, 6, 9]. A radiomics analysis conducted by Zhang et al. using CT-enhanced images from 441 BCa patients revealed the better diagnostic performance of the radiomics model in assessing muscle invasion of BCa [7].

Therefore, to guide the choice of therapeutic treatment option, this study aimed to assess the viability of a radiomics-clinical nomogram for determining the muscle invasion status of BCa.

In the study, we successfully constructed a radiomics-clinical nomogram for preoperative prediction of muscle invasion status in BCa. By comparing the AUC values derived from the ROC curve analysis, the radiomics-clinical nomogram was superior to the other two models (clinical and radiomics) in the training and test cohorts. The same pattern was seen in terms of accuracy and specificity. Finally, the results of the DeLong test demonstrate that the radiomics-clinical nomogram in the training cohort had the best diagnostic performance. The nomogram combining CT radiographic features and clinical risk factors showed better discriminatory and predictive power than the radiographic models in distinguishing MIBC from NMIBC.

The CT examination is time-efficient and has a good sensitivity for detecting calcifications. It provides a visual representation of the bladder wall’s structure, any changes in thickness, and the involvement of adjacent tissues after augmentation [7, 11, 12]. This study included only individuals who had undergone TURBT or cystectomy and had BCA muscle infiltration confirmed. Notably, TURBT pathologic specimens included the muscular layer of the bladder wall, which is characteristic for NMIBC cases [10]. This study examined multiple clinical data points before arriving at statistically significant conclusions. In the clinical model of this study, we included tortuous vessels and histopathological grading. BCas are blood-rich tumors that require nutrient supply for growth, resulting in neoplastic angiogenesis within the lesions [13, 14]. Tumor growth depends on neovascularization within the tumor [14]. Tortuous blood-supplying vessels may be detected in CT-enhanced images of larger tumor lesions [13, 15]. However, the clinical indicator of tortuous vessels was not included in the final radiomics-clinical nomogram after p = 0.084 was found in the multifactorial analysis. It is possible that the tortuous vessel in the CT imaging is a subjective judgment of the radiologist, so it needs to be supported by more cases or multicenter cases.

Clinical staging may commonly be affected by cystoscopic or imaging manifestations, multiple biopsies [16]. This study’s model incorporated pathological tissue grading of the biopsy [17]. Because of the influence of biopsy sampling site, the size of the tissue sampled, and other factors, preoperative and postoperative pathological findings may differ slightly between [5, 18]. Preoperative tissue grading, on the other hand, is useful in predicting the status of BCa muscle layer infiltration. Almost half of newly diagnosed NMIBC are low grade, while the vast majority of MIBC are high-grade [17, 19, 20]. This is consistent with what we discovered. However, it has also been reported in the literature that one-third of NMIBC consist of high-grade BCas [21]. Pathological phenotypes such as grading, staging, and muscle invasion status, according to European Association of Urology (EAU) guidelines, have a significant impact on treatment decisions and prognosis [19, 20]. Li et al. developed a radiomics model for preoperative assessment of histological demarcation of NMIBC based on clinical and radiological data from 169 NMIBC patients [6]. In their study, age and tumor number were included in the model as clinical factors. We also collected age and number of tumors as clinical baseline information, but these features were not finally included in the model.

Radiomics is a technique for extracting and analyzing quantitative imaging features from medical images (e.g., CT, MRI, and other images) [22]. Radiomics is used to develop descriptive and predictive models [14]. In this study, 12 features included morphological features and wavelet filtered features were ultimately extracted from each BCa by radiomics analysis to create radiomics model for distinguishing the infiltration status of bladder intrinsic muscle.

Morphological features included original_shape_Sphericity, original_shape_Elongation, original_shape_LeastAxisLength. The likelihood of bladder cancer muscle invasion was higher when the tumor shape was close to a spherical ellipse. Some scholars predicted the muscle invasion status of bladder cancer based on the tumor-bladder wall contact length (TCL) [23]. Qing Li et al. [23] discovered that an elevated TCL was independently correlated with the muscle infiltration of bladder cancer. Akcay et al. also found that the simultaneous use of VI-RADS criteria and TCL in mpMRI to assess muscular invasion in BCa patients was a successful and highly reproducible method [24]. Tumor size has prognostic and predictive significance in NMIBC, according to EAU guidelines [25]. In this study, an original feature named original_shape_LeastAxisLength was included in the radiomics features. We also included the transverse axial lesion length in CTU images and revealed that the tumor lesion length in MIBC surpassed that of NMIBC. Nevertheless, the difference observed in the training and test groups did not attain statistical significance. This may suggest the superiority of lesion size-related features in radiomics over conventional imaging measurements. The radiomics features contained more wavelet filtered features. The wavelet transform can also gradually convert image information into low- and high-frequency information, which improves local features, increases information content in tumor images, and provides more information about the biological behaviors and heterogeneity of different tumors at multiple scales [26].

The nomogram is a great illustration of how several pertinent pieces of information can be combined to predict a specific endpoint by showcasing the outcomes of a predictive model on a scale. The nomogram can be used to visually determine the patient’s risk for the matching MIBC on the prediction line at the bottom of the nomogram, which can be used to inform the doctor’s decisions about the best course of therapy. The nomogram produced by combining the Radscore and clinical data for the test cohort had an AUC of 0.887, suggesting that the constructed model had sufficient MIBC prediction accuracy. Consequently, the combined nomogram prediction model may aid in directing physicians’ choices.

Our study have several limitations. First, the sample size is quite modest. Second, there is a selection bias, which can restrict the proposed model’s applicability. This study excluded some ineligible cases, for example cases with a small lesion size or the absence of muscle components in the pathological tissue. Additionally, the radiomics model was not externally validated in this study. It is necessary to perform a multicenter validation with more participants.

In summary, to assess muscle invasion in BCa prior surgery, a radiomics-clinical nomogram based on CT images was built in this study. The nomogram showed a high level of diagnostic efficacy and can be used as a guide for BCa prognostic evaluation and tailored treatment.