The goal of the study by Chen M, et al. was to develop a nomogram for diagnosis of gallbladder cancer (GBC) prior to surgical resection [4]. 587 consecutive patients with pathologically verified gallbladder lesions were divided into two groups at random: an internal validation cohort consisting of 287 patients from different hospitals, and a training cohort consisting of 587 individuals [4]. The findings demonstrated that the diagnostic nomogram showed superior performance in detecting GBC with an area under curve (AUC) of 0.91 and 0.89 in internal and external validation cohorts, respectively [4]. It also showed a sensitivity of 91.5% and accuracy of 85% in diagnosing GBC. The nomogram was based on age, CA19.9, and six radiological characteristics [4]. Their study was based on selected cohorts and lacked the multicohort approach. It was retrospective in design which limits its potential impact. Non-transparent selection criteria for the training and validation cohorts with exclusion of comorbid conditions like bile duct anomalies, and lack of consideration of all CT features of GBC were some of the other limitations in their study. In patients with probable GBC, the study by Gupta P, et al. sought to determine the predictive value of computed tomography (CT) texture parameters for histological grade and overall survival [5]. Thirteen of the 38 individuals in the research had no characteristics, and 29 of them had confirmed cancers. In order to distinguish between moderately differentiated and poorly differentiated adenocarcinomas, there were notable variations in the mean and kurtosis at medium textural scales [5]. Kurtosis at medium texture scales was the only texture feature that was found to be significantly correlated with survival [5]. According to their findings, patients with GBC may benefit from using CT texture-based radiomics analysis to predict their histological grade and prognosis [5]. Although their study doesn’t directly involve the use of nomograms, it highlights the role of kurtosis in CT radiomics. Kurtosis represents a variation of pixel intensities that is related to tumor heterogeneity and is often used in developing predictive nomograms. Their study attempted the predict the prognosis of GBC through radiomics which offers a valid advantage. However, the sample size considered was small which limits the generalizability of results. The goal of the study by Xiang F, Liang X, Yang L, Liu X, and Yan S was to create and validate a radiomics signature for gallbladder cancer (GBC) in order to estimate RFS, or recurrence-free survival [6]. 204 GBC patients were included in the study; they were split into cohorts for development and validation [6]. CT scans with contrast added before surgery were used to extract the tumor's radiomic characteristics [6]. Using multivariable Cox regression and significant pathological factors as well as radiomic signatures, a nomogram was created [6]. Patients at high risk with low RFS could be identified using the radiomics signature, which was based on 12 features [6]. Poor differentiation grade, high radiomics score, pT3/4 stage, and pN2 stage were found to be independent risk factors linked to worse RFS by multivariate Cox analysis [6]. At 1, 3, and 5 years, respectively, the nomogram's AUC values of 0.895, 0.935, and 0.907 showed strong prediction ability in estimating RFS [6]. Their study had a moderate sample size and primarily focused on distinguishing between two overlapping entities on imaging, without offering a clear approach for diagnosing them individually. It also did not incorporate other clinical and biochemical parameters, which could have added depth to its findings. Furthermore, the transparency and reproducibility of the deep learning model were not clearly addressed, raising questions about its implementation in different settings. Despite these limitations, the study successfully highlighted the potential of radiomic nomograms in guiding surgical decision-making, showcasing their value in clinical practice. Using a deep learning nomogram model, the study by Zhang W, Wang Q, Liang K, et al. sought to distinguish between GBC and Xanthogranulomatous Cholecystitis (XGC) [7]. 297 individuals with confirmed XGC and GBC were enrolled in the training and internal validation cohort from 2017 to 2021. Using 3-phase merged pictures, the deep learning model Resnet-18 showed remarkable promise in differentiating between XGC and GBC [7]. In the internal validation cohort, the accuracy, precision, and AUC were 0.98, 0.99, and 1.00, whereas in the external validation cohort, they were 0.89, 0.92, and 0.92 [7]. Improved prediction value was demonstrated by the nomogram model that combined clinical features with deep learning prediction scores [7]. According to their study, the nomogram helps surgeons make precise and well-informed surgical decisions for patients with XGC and GBC by accurately differentiating XGC from GBC preoperatively [7]. Their study explored the potential of radiomic nomograms in postoperative gallbladder cancer (GBC) cases, showcasing their strength in assessing tumor extent through pre-contrast CT scans. Combining radiomics with TNM characteristics added depth to the findings, while follow-up data on recurrence-free survival (RFS) at 1, 3, and 5 years highlighted the nomogram’s value in predicting long-term outcomes. However, the study didn’t address the nomogram’s performance in clinical settings, indicating a potential for future research. In order to support preoperative diagnosis and treatment choices for patients with gallbladder cancer (GBC), a study by Jin, Li, Zhang, et al. set out to develop a deep learning-based preoperative clinical radiomics survival prediction model [8]. A high-volume medical center's 168 GBC patients who had preoperative upper abdomen enhanced CT scans between 2011 and 2020 were the subject of the study [8]. Preoperative clinical factors and radiomics data were then combined to create the DeepSurv survival prediction model [8]. The findings demonstrated that in training and testing sets, the DeepSurv model outperformed clinical and nnU-Net-based models in terms of C-index and AUC of 1-, 2-, and 3 year survival prediction, and it was comparable to manual-based models [8]. The nnU-Net-based radiomics DeepSurv model showed a 3 year survival AUC of 0.84 and 0.55 in training and testing sets with equivalent prediction efficiency [8]. According to the study's findings, the DeepSurv model can be used to evaluate and categorize postoperative survival results for specific GBC patients, acting as a guide for preoperative diagnostic and treatment choices [8]. Their study attempted to develop a novel pre-operative prognostic model to predict the risk of malignant gallbladder lesions. By comparing its performance with other AI models, it not only showed promise as a practical tool but also offered insights into how deep learning could be used to improve risk stratification and support clinical decisions. The use of two parameters like the C-index and AUC for comparison of models sets it apart from the previous studies. The purpose of the study by Han S, Liu Y, Li X, et al. was to develop a preoperative nomogram that would use radiomics and clinical factors to differentiate between benign and malignant gallbladder polypoid lesions [9]. After reviewing by 195 general practitioners, radiomic characteristics from three sequences of contrast-enhanced computed tomography were used to create the nomogram [9]. The combined model outperformed the radiomics and clinical models, achieving an AUC of 0.950 [9]. Additionally, in both the training and testing cohorts, the model demonstrated improved sensitivity and specificity [9]. According to the study's findings, a nomogram based on radiomics and clinical characteristics is very useful for differentiating between benign and malignant gall bladder cancers [9]. Despite its strengths, retrospective design of their study introduces potential biases, including selection and recall bias, which may affect the reliability of its findings. Additionally, the use of data from a single center limits the generalizability of the results to broader populations. Furthermore, the exclusive focus of preoperative imaging on polypoidal lesions, without considering other variations of gallbladder cancer could impact the model's applicability in diverse clinical scenarios. Zhuang YY, Feng Y, Kong D, and Guo LL used improved computed tomography (CT) imaging to create a radiomics model for distinguishing benign from malignant gallbladder lesions [10]. Preoperative CT scans were performed on the patients, and radiomic characteristics were retrieved [10]. Groups for testing and training were created from the models [10]. The findings indicated that there were no significant differences between clinical features and radiomics features in the testing group; however, there were substantial disparities in diagnostic accuracy between nomograms and radiomics features in the training group [10]. The testing group's AUCs for the nomogram model were 0.948, the clinical model was 0.904, and the radiomics model was 0.941 [10]. Their study mainly focused on distinguishing between the two types of lesions rather than showing how the tool could be used in diagnosing them individually hindering its clinical utility. However, it did aim to provide a faster and more reliable way to image gallbladder lesions compared to traditional methods, which could help improve preoperative assessments.

Figure 2 illustrates the AUC values of radiographic nomograms in CT diagnosis of GBC across various studies.

AUC values of Radiomic nomograms in CT diagnosis of GBC across various studies

Figures 3, 4 illustrates the application of radiomic nomograms in diagnosis of GBC by Chen, M et al. [4] and Han, Liu, Li, et al. [9] respectively.

Adapted from the study by Chen, et al. [4] radiomic nomogram (radiomic + clinical features) showed the best AUC in diagnosis of GBC compared to clinical factors alone (1), CT value (2), CT values combined with clinical factors (3), and radiological features (4)

Adapted from the study by Han, Liu, Li, et al. [9], the above figure represents a radiomic nomogram developed by combining the Rad score (radiomic features with clinical score)

The goal of the study by Liu X, Liang X, Ruan L, Yan S was to create and verify a nomogram for predicting lymph node metastases in gallbladder cancer (GBC) based on clinical factors and CT radiomics features with a sample size of 353 patients [11]. The least absolute shrinkage and selection operator (LASSO) logistic model was used to create a Radscore [11]. Internal and external validation cohorts were used in the construction and validation of four prediction models [11]. The model that worked best was chosen to create a nomogram [11]. The training cohort, internal validation cohort, and external validation cohort all displayed the highest diagnostic efficiency when using the clinical-radiomics nomogram, which comprised three clinical factors in addition to Radscore [11]. Decision curve analysis (DCA) showed strong clinical utility, and calibration curves showed good discriminating capacity [11]. According to the results, the clinical-radiomics nomogram holds great potential for predicting lymph node metastases in GBC patients prior to surgery [11]. The study, however, has some drawbacks, such as the possibility of bias in the data from two sizable hospitals and the requirement for additional patients from other facilities in order to improve the accuracy of the diagnosis (Table 1).