Research progress of the clinical application of dual-layer spectral computed tomography in gastrointestinal malignancies

Abstract

Gastrointestinal malignancies were the leading causes of cancer-related deaths worldwide in 2022. Owing to the absence of typical clinical symptoms at an early stage, most patients are diagnosed at an advanced stage when the disease has usually progressed to distant metastasis. Biological agents, immune checkpoint inhibitors, and precision radiotherapy have enhanced the efficacy of systemic treatment for locally advanced and metastatic gastrointestinal malignancies. Monitoring tumor malignancy, performing molecular testing of tumors, formulating precise treatment strategies, and evaluating patient prognosis have emerged as key focus areas of research on gastrointestinal malignancies. Dual-layer spectral computed tomography (DLCT) can be used not only to assess tumor histological features but also to quantify them, providing information regarding changes in the tumor microenvironment. Spectral computed tomography is a crucial tool for the initial diagnosis, staging, response evaluation, and follow-up management of patients with gastrointestinal malignancies. This review summarizes the research progress of the clinical application of DLCT in some types of gastrointestinal malignancies, such as esophageal, gastric, and colorectal cancers.

Key Words: Gastrointestinal malignancies; Esophageal cancer; Gastric cancer; Colorectal cancer; Dual-layer spectral detector computed tomography; Radiology

Core Tip: Dual-layer spectral computed tomography (DLCT) achieves true instantaneous dual-energy acquisition through innovative dual-layer detector design, accurately obtaining information for anatomical structure and spectral image. This enables DLCT distinct advantages in detecting gastrointestinal malignancies compared with traditional endoscopic examination. Abundant quantitative parameters and integration of artificial intelligence will further expand the application filed of DLCT in molecular typing, clinical stage, efficacy monitoring and precise management for gastrointestinal malignancy patients.

INTRODUCTION

Esophageal cancer (EC), gastric cancer (GC), and colorectal cancer (CRC) are common types of gastrointestinal malignancies, which were the leading causes of cancer-related deaths worldwide in 2022[1]. Dual-layer spectral computed tomography (DLCT) provides spectral information using a unique double-layer detector, with the top layer capturing lower-energy photons and the bottom layer capturing higher-energy photons. DLCT can achieve simultaneous acquisition with a single X-ray source eliminates temporal and spatial misregistration, favoring superior quantitative accuracy and reproducibility of spectral parameters, but with limited low-energy spectral separation. Dual X-ray source dual-energy computed tomography (CT) has exceptional temporal resolution using two independent source-detector pairs, making it ideal for dynamic imaging, but with cross-scatter and degrade quantitative accuracy, high system complexity and radiation dose. Rapid kilovoltage switching possesses high temporal resolution from ultrafast kilovoltage alternation in a single source, offering good motion artifact correction, but with instantaneous delay and non-simultaneous spectral. Unlike dual X-ray sources or rapid kilovoltage-switching systems, DLCT eliminates the need for altering clinical workflows and selecting dual-energy scanning modes. In addition, it offers the advantages of synchronous projection, automatic exposure control, and without increasing radiation exposure[2]. DLCT provides spectral images such as virtual monoenergetic images (VMIs), iodine maps, virtual non-contrast (VNC) images, and electron density CT maps and quantitative parameters such as iodine concentration (IC), effective atomic number (Zeff), and the slope of spectral HU curves (λHU) through material separation and IC analysis. Compared with conventional CT images, low-kiloelectron volt (keV) VMIs obtained through spectral CT show significantly higher signal-to-noise and contrast-to-noise ratios for patients with EC, GC, or CRC[3-6]. Some studies have shown that low-keV images can significantly improve diagnostic accuracy and liver metastasis of CRC[6,7]. The accuracy of fully automatic segmentation of EC tissues using the 3D Res-UNet algorithm is higher for 40-keV VMIs than for conventional CT images[8]. VNC images reconstructed from the arterial phase (AP) and venous phase (VP) of spectral CT for CRC provide better image quality, and the radiation dose is lower than that required for conventional true non-contrast scans[9]. This review summarizes the clinical applications of DLCT in the diagnosis and prognosis assessment of gastrointestinal cancers, such as EC, GC, and CRC.

EC

DLCT example images of a 72-year-old male EC patient are shown in Figure 1. With the advancement of radiological technology, DLCT has shown potential as a non-invasive method for predicting therapeutic efficacy in EC. With regard to proton therapy for middle-thoracic EC, the stopping power ratio (SPR) derived from DLCT has a dosimetric impact comparable to that of dose calculation based on CT-HU conversion for target coverage and organ-at-risk sparing[10]. The SPR obtained from DLCT scans may be beneficial for planning long-term or complex radiotherapy regimes. Neoadjuvant chemotherapy (NAC) has significantly improved the survival rate of patients with EC, and some drugs have been established as first-line therapy for advanced disease. However, clinical evidence reveals substantial inter-patient variability in the response to NAC. Consequently, developing a novel method for predicting the response of patients with EC to NAC is necessary. DLCT parameters are currently used primarily for assessing the prognosis of patients with EC undergoing NAC. An integrated model constructed based on DLCT-derived parameters, such as arterial enhancement fraction (AEF) and extracellular volume (ECV), and clinical parameters showed good performance in predicting pathological complete response in patients with EC undergoing NAC[11]. This model can be used as a non-invasive method for planning individualized therapeutic strategies for patients with EC. AEF is a quantitative parameter that reflects the tumor blood supply status and evaluates tumor tissue blood perfusion. ECV is a spectral imaging parameter for evaluating tissue fibrosis and alterations in matrix components. Based on the combination of these parameters, the aforementioned model provides more comprehensive information for evaluating therapeutic efficacy. Furthermore, a study reported that an independent predictor combining clinical characteristics (neutrophil-to-lymphocyte ratio and clinical stage) and VP quantitative parameters from DLCT improved the accuracy of predicting the response of patients with EC to NAC[12]. The independent predictors achieved an area under the curve (AUC) of 0.825 in the training set and 0.794 in the validation set. In a study on neoadjuvant therapy for gastroesophageal junction cancer, Graf et al[13] investigated the correlation of normalized IC (NIC) and normalized IC at diagnosis minus IC after neoadjuvant therapy (ΔIC) with tumor regression grade. The AUC value for normalized ΔIC in distinguishing responders from non-responders was 0.95, and that for post-treatment IC was 0.88. These findings suggest that ΔIC after neoadjuvant therapy is a potential imaging biomarker for treatment response. Characteristics of representative research of the clinical application of DLCT in EC depicted in Table 1. Existing studies have used spectral CT mainly for quantifying the clinical treatment response of patients with EC undergoing NAC and proton therapy, which serves as an effective indicator of therapeutic efficacy. However, no definitive conclusions have been reported regarding the application of DLCT, deep learning, and radiomic techniques in assessing molecular changes in preoperative EC tissues and predicting clinical prognosis.

Figure 1 Esophageal cancer and dual-layer spectral computed tomography images. A: Conventional computed tomography image in the arterial phase (AP); B: Effective atomic number in the AP; C: Virtual monochromatic image; D: Iodine concentration computed tomography in the AP.

Table 1 Summary of characteristics of representative research of the clinical application of dual-layer spectral computed tomography in esophageal cancer.

Ref.	Sample size	Cancer type	Main DLCT parameters	Diagnostic performance
Li et al[11]	53 patients	EC	AEF, ECV	The AUC of the combined clinical and DLCT model was 0.893
Wang et al[12]	172 patients	EC	Zeff-VP and NIC-VP	The spectral CT and clinical model yielded the highest AUC of 0.825
Graf et al[13]	62 patients	EC	NIC, normalized ΔIC	The normalized ΔIC yielded the highest AUC of 0.95; ICafter NAC achieved an AUC of 0.88

DLCT: Dual-layer spectral computed tomography; CT: Computed tomography; EC: Esophageal cancer; AEF: Arterial enhancement fraction; ECV: Extracellular volume; AUC: Area under curve; Zeff: Effective atomic number; VP: Venous phase; IC: Iodine concentration; NIC: Normalized iodine concentration; NAC: Neoadjuvant chemotherapy; ΔIC: Iodine concentration at diagnosis-iodine concentration after normalized iodine concentration.

GC

Diagnosis of GC

DLCT example images of a 74-year-old female GC patient are shown in Figure 2. Accurate assessment of TP53 mutations and Ki-67 levels can facilitate the selection of appropriate treatment strategies for GC. A study reported that a model incorporating clinical features and DLCT-derived quantitative parameters showed significantly higher performance in predicting TP53 expression compared to clinical and conventional CT-based models[14]. Several studies have indicated that quantitative parameters derived from spectral CT can differentiate between varying levels of Ki-67 in GC, indicating that Ki-67 status is positively correlated with IC and NIC in the VP[15,16]. The therapeutic benefits of immunotherapy vary depending on the microsatellite instability (MSI) status of patients. Consequently, developing novel techniques for assessing the MSI status of patients with GC is necessary. Zhu et al[17] reported that NIC-_VP derived from spectral CT exhibited the highest predictive efficacy in discriminating MSI status. A predictive model incorporating DLCT parameters (NIC-_VP, Zeff-_VP, and λHU-_VP), tumor location, and lymph node stage evaluated from DLCT images (N-_CT stage) showed superior accuracy in predicting the MSI status of patients with GC. This model may serve as a valuable tool for postoperative risk stratification.

Figure 2 Gastric cancer and dual-layer spectral computed tomography images. A: Conventional computed tomography image in the arterial phase (AP); B: Effective atomic number in the AP; C: Virtual monochromatic image; D: Iodine concentration in the AP.

Accurate clinical staging and pathological typing are of great significance in determining clinical treatment strategies and stratifying risk for GC patients. Zhang et al[18] established a nomogram incorporating clinical features and spectral CT–derived quantitative parameters for non-invasive Lauren classification in patients with GC preoperatively. The AUC value of the nomogram model was 0.841, with a sensitivity of 68.4%, specificity of 85.3%, and accuracy of 76.4%. Li et al[19] reported that IC, NIC, Zeff, normalized Zeff (NZeff), and attenuation of AP and VP were significantly lower in the benign group than in the malignant gastric wall thickening group. The AUC values of IC, NIC, and attenuation in the VP were 0.864, 0.86, and 0.840, respectively, and that of a newly developed model for differentiating benign from malignant gastric wall thickening was 0.884. These findings indicate that spectral CT–derived quantitative parameters are valuable in distinguishing between benign and malignant gastric lesions and may help avoid unnecessary endoscopic examinations in non-malignant patients. Zeng et al[20] enrolled 148 patients with GC who underwent DLCT before surgery (diagnosis was confirmed through histopathological analysis). In both training and test sets, a model integrating DLCT-derived parameters with radiomic features (RFs) showed superior performance in differentiating T₄ from non-T₄ stages compared to subjective evaluation, clinical models, spectral parameters alone, and conventional CT models. The AUC value was 0.906 in the test set and 0.873 in the training set. Independent predictors[21], including CT attenuation on 70 keV-_AP images, ED-_VP, and clustered features, and a model[22] combining delayed-phase IC, NIC-_AP, and ECV showed improved predictive performance for lymph node metastasis in GC.

Prediction of treatment response in GC

A nomogram model incorporating spectral parameters (λHU-_VP and HU values from 40 keV-_VP VMIs) and visceral fat area showed better performance in predicting postoperative complications in patients with GC[23]. Li et al[24] examined 65 patients who underwent spectral CT with three-phase enhancement, magnetic resonance imaging (MRI), standard NAC, and radical gastrectomy. NIC-_DP and apparent diffusion coefficient (ADC) were significantly correlated with tumor regression grade, yielding AUC values of 0.674 and 0.673, respectively. A model incorporating NIC and ADC significantly improved the AUC value to 0.770. These results suggest that integrating spectral CT–derived parameters with functional MRI may provide multimodal biomarkers for predicting the early response to NAC in GC patients. Radiomic technology has expanded the scope of research on DLCT and enhanced its application in evaluating treatment efficacy in patients with tumors. For instance, a model integrating clinical features with DLCT-derived parameters, such as ECV and N-_CT stage, and RFs derived from GC tumors and peritumoral fat showed superior predictive performance and diagnostic value for lymph node metastasis compared to a clinical DLCT-based model and a radiomic model[25]. Characteristics of representative research of the clinical application of DLCT in GC depicted in Table 2.

Table 2 Summary of characteristics of representative research of the clinical application of dual-layer spectral computed tomography in gastric cancer.

Ref.	Sample size	Cancer type	Main DLCT parameters	Diagnostic performance
Wu et al[14]	568 patients	GC	CT 40keV, NIC-VP, NZeff-VP	Clinical-DLCT model scoring system enables noninvasively, cost-effectively and rapidly predict TP53 expression
Mao et al[15]	108 patients	GC	Zeff-VP and NIC-VP	The Zeff-VP and NIC-VP (AUC: 0.835 and 0.805) showed better performance in discriminating the Ki-67 status
Du et al[16]	72 patients	GC	NIC, IC and iodine-no-water concentration	A positive correlation between Ki-67 expression levels and IC, NIC, and iodine-no-water concentration in the VP
Zhu et al[17]	264 patients	GC	NIC-VP, Zeff-VP, λHU-VP	The model including DLCT parameters, tumor location and N-CT stage in predicting the MSI status of GC achieved a high prediction efficacy in the validation set, with AUC of 0.879
Zhang et al[18]	72 patients	GC	NIC-AP and λHU-DP	The nomogram based on these indicators (Gender, NIC-AP and λHU-DP) for Lauren classification produced the best performance with an AUC of 0.841
Li et al[19]	58 patients	GC	IC-VP, NIC-VP, and attenuation in the VP	The combination of these factors (IC, NIC, and attenuation in the VP) and gastric wall thickness for differentiation of benign and malignant gastric wall thickening had an AUC of 0.884
Luo et al[21]	55 patients	Lymph nodes of GC	CT attenuation on 70 keV-AP images, ED-VP, and clustered features	These combination predictors (CT attenuation on 70keV-AP images, ED-VP, and clustered features) in diagnosing of metastatic lymph nodes of GC had AUC of 0.855 and 0.907 in the training and validation sets
Zhang et al[22]	70 patients	Lymph nodes of GC	IC-DP, NIC-AP, and ECV	The diagnostic efficacy of ECV% for predicting Lymph nodes metastases of GC was higher than that of other parameters in training and test sets (AUC = 0.823 and 0.803). Model 3 (spectral CT parameters and ECV%) for predicting lymph nodes metastases of GC demonstrated significantly higher diagnostic efficacy than other models in training and test sets (AUC = 0.858 and 0.881)
Tan et al[23]	101 patients	GC	λHU-VP and HU values from 40 keV-VP VMIs	The nomogram based on two tumor spectral parameters (λHU-VP, 40 keV-VP VMIs) and VFA yielded an AUC of 0.89 predicting the POCs of GC patients
Li et al[24]	65 patients	GC	NIC-DP	A model incorporating NIC-DP and ADC significantly improved the AUC value to 0.770

DLCT: Dual-layer spectral computed tomography; GC: Gastric cancer; keV: Kiloelectron volt; AUC: Area under curve; Zeff: Effective atomic number; Nzeff: Normalized effective atomic number; AP: Arterial phase; VP: Venous phase; DP: Delayed phases; ED: Electron density; IC: Iodine concentration; NIC: Normalized iodine concentration; MSI: Microsatellite instability; λHU: Slope of spectral HU curves; VMIs: Virtual monoenergetic images; VFA: Visceral fat area; POCs: Postoperative complications; ECV: Extracellular volume.

Overall, DLCT-derived quantitative parameters can help design personalized treatment strategies and predict molecular expression, clinical stage, and pathological type in patients with GC. However, existing evidence remains preliminary, warranting further research.

CRC

DLCT example images of a 57-year-old male CRC patient are shown in Figure 3. Spectral cleansing based on DLCT can reduce electronically tagged stool cleansing artifacts in CT colonography (CTC)[26]. A study showed that VMIs generated from DLCT significantly improved CTC image quality and enhanced the effectiveness of electronically cleansing[27]. This approach overcomes the challenges associated with onerous bowel preparation for colon cancer screening and poor tolerance of patients to colon examination to some extent.

Figure 3 Colorectal cancer and dual-layer spectral computed tomography images. A: Conventional computed tomography image in the arterial phase (AP); B: Effective atomic number in the AP; C: Virtual monochromatic image; D: Iodine concentration in the AP.

Diagnosis of CRC

Tumor stages and histological features of pathological tissues are crucial determinants of prognosis in patients with CRC. A study on 80 patients with local colonic wall thickening (LCWT) who underwent contrast-enhanced CT and colonoscopy showed that the IC and NIC of patients with non-pathological LCWT were significantly lower than those of patients with pathological LCWT[28]. The sensitivity and specificity of IC for diagnosing colon tumors were 91.5% and 75.8%, respectively, whereas those of NIC were 85.1% and 84.8%, respectively. The DLCT-derived parameters IC and NIC can reflect blood flow and vascular distribution in tumors. The formation of new blood vessels is related to tumor growth and invasion. Compared with IC, NIC can more effectively alleviate the impact of individual circulatory changes on iodine content within tumors, thereby reflecting blood supply at the lesion site more accurately. ECV has been identified as a useful parameter for predicting the T stage in CRC, with an AUC value of 0.919 in a training set and 0.892 in an external validation set, enabling differentiation between the T₁/T₂ and T₃ stages of CRC[29]. In addition, Zeff, NIC-_AP, and λHU-_AP can successfully distinguish between different stages and grades of CRC[30]. Attenuation at 40 keV, IC, and NIC during AP and attenuation at 40 keV and IC during VP show significant differences across histological differentiation grades in patients with CRC, with the parameters from AP exhibiting better diagnostic performance. Different tissues, organs, and lesions exhibit distinct λHU energy spectrum curves, and Zeff reflects the nature of the interaction between materials and ionizing radiation. Therefore, Zeff and λHU values derived from DLCT vary among patients with different stages and grades of CRC.

Tumor deposits (TDs), perineural invasion (PNI), and MSI are closely associated with clinical stage, increased tumor invasiveness, prognosis, and treatment response. Rectal cancer (RC) with PNI is associated with increased recurrence and reduced survival rates. A model combining tumor N stage and attenuation at 40 keV showed optimal efficacy in assessing the PNI status in RC. The predictive ability of most quantitative parameters derived from spectral CT is superior to that of parameters derived from conventional CT[31]. However, Chen et al[32] reported inconsistent findings. They found that spectral CT–derived parameters did not significantly vary with the PNI status. Another study showed that a clinical-radiomic model constructed based on iodine maps showed higher efficacy in predicting the preoperative MSI status of patients with CRC compared to a clinical model[33]. Accurate assessment of TD status is achieved through pathological examination following surgical resection. Feng et al[34] retrospectively analyzed 264 patients with pathologically confirmed CRC and found that a model incorporating tumor T stage, lymph node status, NIC from spectral CT, and RFs from iodine maps showed strong efficacy in predicting the preoperative TD status in patients with CRC. The model achieved an AUC value of 0.926 in the training set, 0.881 in the test set, and 0.887 in the external validation set, outperforming both the clinical and radiomic models.

The lymph node metastasis status determines the surgical strategy, indications for adjuvant chemotherapy, and risks of recurrence in patients with CRC. In T₁/T₂ RC, a combination of spectral CT–derived NZeff-_VP and the short-axis diameter of lymph nodes has shown high efficacy in identifying metastatic lymph nodes[35]. Patients with CRC with negative lymphovascular tumor thrombus have a significantly better prognosis than those with positive lymphovascular tumor thrombus. The NIC-_VP value helps predict the lymphovascular tumor thrombus status in patients with CRC before surgery[36]. HU values at 40 keV and 100 keV, Zeff, IC, and λHU significantly vary depending on CRC grade and the presence of lymphovascular invasion[32]. A retrospective study reported that a model integrating clinical features and RFs from both conventional and spectral CT images showed good calibration and significant net benefit in detecting lymph node metastasis in CRC, with AUC values of 0.879 in the training set and 0.824 in the validation set[37]. Postoperative complications in colon cancer can significantly impact patient prognosis and reduce overall survival rates.

Prediction of treatment response in CRC

Accurate prediction of treatment-related complications and prognostic evaluation are crucial for guiding individualized treatment strategies. A study reported that IC can be used to evaluate the response of patients with RC to radiochemotherapy[38]. In addition, IC showed significant changes after radiochemotherapy, and Spearman’s P value of the absolute IC difference was significant. Several studies have shown that DLCT-derived parameters from VP and clinical features of CRC can be used to predict the risk of postoperative complications and distant metastasis or recurrence in CRC[39-42]. These spectral parameters are closely related to tumor perfusion in VP. This relationship may be attributed to the delayed enhancement pattern of colon cancer, as VP characteristics show a stronger correlation with microvascular enhancement and consequently better reflect the biological features of tumors. Therefore, the application of DLCT can not only help clinicians understand pathological features such as histological type, clinical stage, degree of differentiation, tumor markers, TDs, PNI, and MSI, but also enable non-invasive assessment of the risk of postoperative complications in patients with CRC, thereby improving diagnostic accuracy and guiding personalized treatment plans to improve long-term prognosis. However, whether preoperative DLCT–derived parameters can effectively predict the clinical response to NAC and serve as a reliable indicator for prognosis assessment remains unclear. A model incorporating clinical features, spectral CT–derived quantitative parameters, and radiomic data showed excellent efficacy in predicting tumor heterogeneity in CRC, significantly improving prediction accuracy and outperforming single-factor-based models. Fan et al[43] showed that 3D quantitative parameters derived from DLCT achieved higher interobserver accuracy than two-dimensional parameters in CRC. They established a multidimensional radiological–angiogenic–clinical pathological model, which showed an AUC value of 0.95 in the training set and 0.93 in the validation set. Characteristics of representative research of the clinical application of DLCT in CRC depicted in Table 3.

Table 3 Summary of characteristics of representative research of the clinical application of dual-layer spectral computed tomography in colorectal cancer.

Ref.	Sample size	Cancer type	Main DLCT parameters	Diagnostic performance
Wang et al[28]	80 patients	CRC	IC and NIC	The AUC of IC for diagnosing colon tumors was 0.837, with a sensitivity of 91.5% and a specificity of 75.8%. The AUC of NIC for diagnosing colon tumors was 0.899, with a sensitivity and specificity of 85.1% and 84.8%
Sun et al[29]	165 patients	CRC	ECV	ECV had diagnostic efficacy for CRC pT staging in both the training and external validation sets (AUC = 0.919 and 0.892)
Chen et al[30]	131 patients	CRC	Zeff, NIC-AP, NIC-VP, and λHU-AP	The AUCs of Zeff, NIC-AP, NIC-VP, and λHU-AP for distinguishing different stages of CRC were 0.83, 0.80, 0.79, 0.86, and 0.68, respectively. The AUCs of Zeff, NIC-AP, NIC-VP, and λHU-AP for distinguishing different grades of CRC were 0.83, 0.80, 0.79, 0.86, and 0.68, respectively
Lu et al[31]	62 patients	CRC	40 keV attenuation	The nomogram incorporating these two predictors (N-CT stage and 40 keV attenuation) exhibited the best efficacy in the preoperative assessment of PNI status in RC, with an AUC of 0.885
Chen et al[32]	106 patients	CRC	40 keV-VP, 100 keV-VP, Zeff-VP, IC-VP and λHU-VP	The AUCs of 40 keV-VP, 100 keV-VP, Zeff-VP, IC-VP, and λHU-VP in distinguishing LVI status of CRC were 0.688, 0.644, 0.688, 0.703, 0.677, respectively. Spectral CT derived parameters did not significantly vary with the PNI status
Chen et al[33]	255 patients	CRC	IC-AP, NIC-VP	A clinical–radiomic model constructed based on iodine maps showed higher efficacy in predicting the preoperative MSI status
Liu et al[35]	42 patients	Lymph nodes of CRC	NZeff-VP	After combining NZeff and the short-axis diameter, the AUC (0.966) in diagnosing metastatic Lymph nodes in CRC patients was the highest with sensitivity of 100% and specificity of 87.7%
Sauter et al[38]	11 patients	CRC	Absolute IC difference	IC values decreased significantly after RCT. The absolute IC difference and the absolute ADC (both before and after RCT) is high and significant
Peng et al[39]	222 patients	CRC	VEF, λHU-VP and 1/NIC-VP	The clinical-spectral model in predicting VEDM in CRC following surgery achieved further improved AUC of 0.887
Yang et al[40]	100 patients	CRC	IC-VP, λHU and CT attenuation 40 keV	The nomogram based on spectral CT parameters, CEA, and CA199 demonstrated high discriminative ability, with AUC of 0.9078 in the training set and 0.9502 in the internal validation set
Liu et al[41]	134 patients	CRC	NIC-VP, λHU-VP	The combined indicator integrating NIC-VP, λHU-VP and CEA achieved the best diagnostic performance (AUC = 0.900) in predicting prognosis in RC
Tan et al[42]	85 patients	CRC	40 keV-VP VMIs and VFA	The combined model based on predictors (40keV-VP VMIs and VFA) in predicting POCs in colon cancer produced an AUC of 0.84, with a sensitivity of 77.8% and specificity of 87.9%

DLCT: Dual-layer spectral computed tomography; CRC: Colorectal cancer; ECV: Extracellular volume; AUC: Area under curve; Zeff: Effective atomic number; NZeff: Normalized Zeff; IC: Iodine concentration; NIC: Normalized iodine concentration; keV: Kiloelectron volt; AP: Arterial phase; VP: Venous phase; DP: Delayed phases; λHU: Slope of spectral HU curves; PNI: Perineural invasion; LVI: Lymph vascular invasion; RCT: Radiochemotherapy; MSI: Microsatellite instability; VEF: Venous enhancement fraction; VEDM: Very early distant metastasis; VMIs:Virtual monoenergetic images; VFA: Visceral fat area; CEA: Carcinoembryonic antigen; POCs: Postoperative complications; RC: Rectum cancer.

CONCLUSION

Radiomics and deep learning have demonstrated excellent performance in image data processing, significantly enhancing the specificity of cancer detection in pathological analysis and the accuracy of prognosis assessment. Different levels of VMIs derived from DLCT help extract multidimensional information from the same scan, which is beneficial for radiomic analysis. Xu et al[44] compared the reproducibility of RFs among ground-truth VMIs, conventional single-energy CT scans, and synthetic VMIs generated from wavelet loss–enhanced cycle generative adversarial networks (CycleGANs). Deep learning methods can reduce the significant impact of material decomposition on VMIs. CycleGANs can improve the robustness and repeatability of RFs for EC. The combination of DLCT and radiomics also has the potential to enhance the accuracy and practical value of predicting some clinical pathological conditions and lymph nodes metastasis in GC and CRC. However, currently, there are still limited studies combining DLCT with deep learning to predict the pathological characteristics and prognosis of gastrointestinal malignancies. In addition, differences in image data quality, resolution, and imaging equipment can affect feature extraction performance, resulting in reduced generalizability on datasets. Although it faces challenges such as data, generalization, and interpretability, its application prospects in oncology, cardiovascular diseases, neuroscience and other fields are extremely broad, representing the core direction of future imaging diagnosis development.

Compared with conventional CT scans, low-keV VMIs derived from DLCT show higher image quality, improving the assessment of patients with gastrointestinal malignancies. Spectral CT images and their associated parameters can reflect tissue composition, tumor angiogenesis, and blood supply. In existing studies, the delineation of target regions remains highly subjective and is limited by the lack of multi-center validation, both of which can compromise the accuracy of experimental outcomes. Furthermore, the differences in imaging analysis of malignant gastrointestinal tumors across various DLCT systems remain unclear, and the advantages of spectral CT have not been comprehensively reported. Therefore, further exploration and research on gastrointestinal malignancy DLCT integrating multi-center, multi-device and artificial intelligence is still required, although an amount of clinical data on EC, GC, and CRC using DLCT has emerged. Overall, DLCT can help determine lesion characteristics, thereby assisting clinicians in formulating individualized treatment plans.