Skip to main content
Download PDF

Development and validation of a multi-modal ultrasomics model to predict response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer

BMC Medical Imaging volume 24, Article number: 65 (2024) Cite this article

Abstract

Objectives

To assess the performance of multi-modal ultrasomics model to predict efficacy to neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC) and compare with the clinical model.

Materials and methods

This study retrospectively included 106 patients with LARC who underwent total mesorectal excision after nCRT between April 2018 and April 2023 at our hospital, randomly divided into a training set of 74 and a validation set of 32 in a 7: 3 ratios. Ultrasomics features were extracted from the tumors’ region of interest of B-mode ultrasound (BUS) and contrast-enhanced ultrasound (CEUS) images based on PyRadiomics. Mann-Whitney U test, spearman, and least absolute shrinkage and selection operator algorithms were utilized to reduce features dimension. Five models were built with ultrasomics and clinical analysis using multilayer perceptron neural network classifier based on python. Including BUS, CEUS, Combined_1, Combined_2 and Clinical models. The diagnostic performance of models was assessed with the area under the curve (AUC) of the receiver operating characteristic. The DeLong testing algorithm was utilized to compare the models’ overall performance.

Results

The AUC (95% confidence interval [CI]) of the five models in the validation cohort were as follows: BUS 0.675 (95%CI: 0.481–0.868), CEUS 0.821 (95%CI: 0.660–0.983), Combined_1 0.829 (95%CI: 0.673–0.985), Combined_2 0.893 (95%CI: 0.780-1.000), and Clinical 0.690 (95%CI: 0.509–0.872). The Combined_2 model was the best in the overall prediction performance, showed significantly better compared to the Clinical model after DeLong testing (P < 0.01). Both univariate and multivariate logistic regression analyses showed that age (P < 0.01) and clinical stage (P < 0.01) could be an independent predictor of efficacy after nCRT in patients with LARC.

Conclusion

The ultrasomics model had better diagnostic performance to predict efficacy to nCRT in patients with LARC than the Clinical model.

Peer Review reports

Introduction

Rectal cancer, one of the most common malignant tumors globally, has the second highest mortality rate among malignant tumors according to the global epidemiological survey in 2020 [1]. Therefore, prompt diagnosis and timely intervention are crucial in mitigating the incidence of relapse and fatality rates associated with rectal cancer. Undergo the total mesorectal excision (TME) after neoadjuvant chemoradiotherapy (nCRT) is the standard treatment for patients with locally advanced rectal cancer (LARC), which induce tumors downsizing and downstaging [2]. However, there are individual differences in tumor response after nCRT, with some patients responding well, and approximately 15–27% of patients can attain a pathological complete response (pCR) after nCRT [3]. About 30–40% of patients have poor response after nCRT, and even a few patients experience tumor progression [4]. Early identification of patients with good response to nCRT can guide treatment strategies to improve quality of life and prognosis. In contrast, patients with poor response to nCRT not only fail to achieve effective control the tumor, but also suffer from nCRT toxicity damage, which can lead to worse quality of life and prognosis [5]. Therefore, it is crucial to explore different predictive methods for early identification of LARC patients with good response after nCRT, which will guide personalized treatment and clinical surgical decision-making for patients.

Recently, multiple studies had reported the predictive performance of CT and MRI radiomics for the efficacy after nCRT in patients with LARC [6,7,8], while fewer reports of ultrasomics for the efficacy of nCRT in patients with LARC. However, comparing to CT and MRI examinations, ultrasound examination has advantages such as no radiation and short duration. If more diagnostic information can be mined by ultrasomics from ultrasound images to identify LARC patients with good response after nCRT, thus guide patient treatment, this may provide additional value for future multimodal radio studies in rectal cancer. Existing evidence suggested that ultrasomics had a good performance in predicting lymphovascular invasion in rectal cancer [9] and the response of nCRT in breast cancer [10,11,12]. It had also shown a good diagnostic performance in tumor classification, staging, and differentiation between benign and malignant tumors [13, 14].

The objective of this research is to evaluate the performance of multi-modal ultrasomics models to predict the efficacy in patients with LARC after nCRT and compare with the Clinical model. Univariate and multivariate logistic regression analyses are used to select independent predictors of good response after nCRT in patients with LARC.

Materials and methods

This retrospective study was conducted with approval from the Medical Ethics Committee of First Affiliated Hospital of Guangxi Medical University (No.2023-E276-01). The requirement of informed consent was waived by the Ethics Committee of First Affiliated Hospital of Guangxi Medical University owing to the retrospective nature of the study. This study was performed in accordance with the ethical standards of the institutional and national research committees as well as the Helsinki Declaration.

Population

We retrospectively included patients with LARC who underwent TME after nCRT in our hospital from April 2018 to April 2023. Inclusion criteria: (a) patients with LARC received rectal ultrasound examination and had complete B-mode ultrasound (BUS) and contrast-enhanced ultrasound (CEUS) images before nCRT. (b) patients with LARC had completed standard nCRT. (c) complete postoperative pathological data after TME. (d) no prior radiotherapy, chemotherapy or immunotherapy in the past. Exclusion criteria: (a) without TME after nCRT; (b) unable to cooperate to complete ultrasound examination; (c) only with BUS examination without CEUS examination before nCRT; (d) patients with incomplete clinical information. This study design illustrates in Fig. 1.

Fig. 1
figure 1

Flowchart of this study

Ultrasomics features extracted from the region of interest of tumors based on BUS and CEUS images, respectively. Mann-Whitney U test, spearman, and LASSO were utilized to reduce feature’s dimension. Five models were built based on ultrasomics and clinical analysis using MLP classifier

BUS: B-mode ultrasound; CEUS: contrast-enhanced ultrasound; LASSO: least absolute shrinkage and selection operator; MLP: multilayer perceptron neural network

Full size image

Histopathological reference standard

Based on the tumor regression grade (TRG) standard of the eighth edition guide of the AJCC [15]. Two experienced pathologists independently double-blindly classified all patients into four levels and resolved issues by discussion when there was disagreement. TRG was as follows: TRG0 indicates a pathological complete response (indicating no viable tumor cells are present). TRG1 denotes a near pathological complete response (indicating the presence of single or small groups of tumor cells). TRG2 indicates a minimal response (representing the residual cancer outgrown by fibrosis). TRG3 indicates an inferior response (representing minimal or no tumor cells were killed).

This study clustered patients into two cohorts: (1) patients with TRG0 − 1 were defined as good response; and (2) patients with TRG2 − 3 were defined as poor response [16].

Clinical baseline data

Clinical baseline data of patients before nCRT included age, gender, tumor location, tumor size, clinical stage, T stage, N stage, distant metastasis, CEA, TRG, CA199 and CA242.

Ultrasound examination protocol

Rectal ultrasound examination was performed using ultrasound diagnostic machines with LOGIQ E9 (IC-5-9-D probe, frequency 5–9 MHz) from General Electric Company in the USA and Mylab Class C (TRT33 probe, convex array mode frequency 3–9 MHz, linear array mode frequency 4–13 MHz) from Italian Esaote. The maximum cross-section of the lesion was first found in grey scale mode and then switched to contrast mode. The ultrasound contrast agent was SonoVue, which was injected with 5 ml of physiological saline before use and configured as a suspension. 2.4 ml SonoVue was quickly injected through the median cubital vein, followed by 5 ml of saline to flush the tube. Synchronous timing and recording, and continuous observation for 3 min. The images were stored and preserved in DICOM format. After offline, the maximum cross-section of the lesion in the early phase of contrast (within 30 s after injection of contrast agent) with the best image quality was selected for subsequent analysis.

Ultrasomics analysis

A radiologist (Q.Q.) with eight years of experience in rectal ultrasound used ITK-SNAP (version4.0.1; http://www.itksnap.org) to manually segment the region of interest (ROI) of tumors from the BUS and CEUS images, respectively. A senior radiologist confirmed all segmentation masks and any conflicting opinions resolved by discussion. The maximum ROI cross section was extracted from the BUS images and saved in PNG format for further feature extraction with the ResNet101 which was a deep learning model. All the pictures were resampled to a voxel dimension of 1 × 1 × 1 mm. The voxel intensity values were quantized using a bin size of 25 HU [17].

The ultrasomics features were extracted from the BUS and CEUS pictures utilizing PyRadiomics (version 3.0) [18], independently. Which was an open-source software. All features were derived from the initial image and processed images. 1239 features were obtained in all, including 7 types: (1) 17 features of Shape; (2) 247 features of First Order; (3) 182 features for Gray Level Dependence Matrix; (4) 208 features of Gray Level Size Zone Matrix; (5) 65 features of Neighborhood Gray Tone Difference Matrix; (6) 208 features of Gray Level Run Length Matrix; (7) 312 features of Gray Level Co-occurrence Matrix [19]. Additionally, we utilized the pre-trained ResNet101 that a deep learning model to extract 2038 features from the maximum ROI cross-sectional images. After features compression processing, 32 deep learning features were eventually included. All features were standardized using Z-score algorithm to transform them into a uniform measure for comparison.

To avoid overfitting, we employed the Mann-Whitney U-test to choose ultrasomics features with significant distinctions between good response and poor response cohorts. Then spearman’s correlation coefficients were calculated and reserved the features that correlation coefficient > 0.9. To minimize redundancy among radiomic features, we utilized the LASSO (least absolute shrinkage and selection operator) algorithm to reduce features dimension and ten-fold cross validation to select the most relevant features of good response. Built four ultrasomics models using a multilayer perceptron neural network (MLP) classifier based on python, which was an artificial neural network consisting of numerous interconnected neural nodes or layers, where each layer was fully connected to the next layer. Each layer had different weights and was trained using backpropagation. The input layer was responsible for receiving input data, while the output layer produced the final classifications output, and the hidden layers performed different non-linear transformations on the input data. We also used ten-fold cross validation to train the model and selected the optimal parameters to improve model stability. Models were as follows: the BUS model was formed based on only BUS features; the CEUS model was formed based on only CEUS features; the Combined_1 model was created based on the integration of BUS and CEUS features; and the Combined_2 model was created based on the integration of BUS, CEUS, and deep learning features. Additionally, we created a clinical model based on clinical factors.

Statistical analyses

SPSS software (version 23.0) was employed to process clinical parameters, univariate and multivariate logistic regression analyses were used to select parameters with statistically significant differences between groups. Features selection and models construction performed with Python software (version 3.7). Count data was exhibited with percentage and differences between groups were tested using chi-square test. Metrological data obeying normal distribution was expressed as mean ± standard deviation, and t-test was used for intergroup comparisons; Mann-Whitney U test was used for intergroup comparisons of metrological data obeying skewed distribution. Two-sided P values less than 0.05 were considered statistically significant differences. The parameters of diagnostic performance, comprised the area under the curve (AUC 95% confidence interval [CI]), positive predictive value (PPV), and negative predictive value (NPV), specificity, sensitivity, accuracy. The DeLong testing algorithm was utilized to evaluate the overall prediction performance of various models [20].

Results

Patient characteristics

This study included 106 patients ultimately, including 51 patients with good response (48.1%) and 55 patients with poor response (51.9%). Patients were randomly divided into a training set 74 and a validation set 32 at 7:3 ratios (Fig. 2), including 72 male (68%) and 34 female (32%) patients, aged 19–78 (mean 54.3 ± 10.0) years.

Fig. 2
figure 2

Inclusion and exclusion criteria of patients

nCRT: neoadjuvant chemoradiotherapy; TME: total mesorectal excision; BUS: B-mode ultrasound; CEUS: contrast-enhanced ultrasound

Full size image

All clinical pathological parameters, only CEA showed a statistically significant difference between the training set and validation set (P < 0.01) (Table 1). Both univariate and multivariate logistic regression analyses showed that age (P < 0.01) and clinical stage (P < 0.01) could be an independent predictor of good response after nCRT in patients with LARC (Table 2).

Table 1 Clinicopathological parameters between training and validation cohorts
Full size table
Table 2 Univariable and multivariable logistic analysis of the clinical parameters
Full size table

Ultrasomics analysis

After z-score normalization, the BUS model had 1237 features; the CEUS model had 1237 features; the Combined_1 model had 2474 features; and the Combined_2 model had 2506 features. After Mann-Whitney U test, the BUS model had 10 features; the CEUS model had 44 features; the Combined_1 model had 50 features; and the Combined_2 model had 52 features. After spearman correlation analysis, the BUS model had 6 features; the CEUS model had 16 features; the Combined_1 model had 20 features; and the Combined_2 model had 22 features. The final step utilized the LASSO method for features selection, and ultimately, BUS, CEUS, Combined_1, and Combined_2 models incorporated 4, 7, 11, and 12 features respectively to build the ultrasomics models. Table 3 provides information on the included features and corresponds coefficients in every model.

Table 3 Features in every model
Full size table

In the training cohort, sensitivity and specificity were 73.0% and 70.3% for the BUS model, 70.3% and 64.9% for the CEUS model, 67.6% and 89.2% for the Combined_1 model, 75.7% and 83.8% for the Combined_2 model, respectively. In the validation cohort, sensitivity and specificity were 57.1% and 77.8% for the BUS model, 64.3% and 94.4% for the CEUS model, 92.9% and 66.7% for the Combined_1 model, 78.6% and 88.9% for the Combined_2 model, respectively (Table 4). In the training cohort, the AUC of BUS, CEUS, Combined_1, and Combined_2 models to predict good response were 0.751 (95%CI, 0.640–0.862), 0.714 (95%CI, 0.598–0.831), 0.833 (95%CI, 0.740–0.927) and 0.844 (95%CI, 0.757–0.932), respectively. And 0.675 (95%CI, 0.481–0.868), 0.821 (95%CI, 0.660–0.983), 0.829 (95%CI, 0.673–0.985), and 0.893 (95%CI, 0.780-1.000), respectively, in the validation cohort (Table 5; Fig. 3). The outcomes of DeLong testing demonstrated that the prediction performance of the Combined_2 and Combined_1 models regardless of whether it was the training cohort (P = 0.774) or the validation cohort (P = 0.140) had no significant distinction in statistics (Table 5). Therefore, we would compare these two models with the Clinical model.

Table 4 Parameters of diagnostic performance
Full size table
Table 5 DeLong testing between ultrasomics models
Full size table
Fig. 3
figure 3

Diagnostic performance of ultrasomics models. (A) The ROC curves of four ultrasomics models in training cohort. (B) The ROC curves of four ultrasomics models in validation cohort. BUS: B-mode ultrasound; CEUS: contrast-enhanced ultrasound; Combined_1: the integration of BUS and CEUS; Combined_2: the integration of BUS: CEUS: and deep learning; AUC: area under the curve

Full size image

Comparison of ultrasomics models and clinical model

The AUC for the Combined_1 and Combined_2 models was higher than the Clinical model regardless of whether it was the training group (0.833 [95%CI, 0.740–0.927] for the Combined_1 model and 0.844 [95%CI, 0.757–0.932] for the Combined_2 model vs. 0.680 [95%CI, 0.569–0.792]) or the validation group (0.829 [95%CI, 0.673–0.985] for the Combined_1 model and 0.893 [95%CI, 0.780-1.000] for the Combined_2 model vs. 0.690 [95%CI, 0.509–0.872]) (Table 6; Fig. 4). The results of DeLong testing demonstrated that the AUC of the Combined_2 model was greater than that of the Clinical model regardless of whether it was the training cohort (P = 0.007) or the validation cohort (P = 0.006) (Table 6). The AUC, accuracy, specificity, and PPV of the Combined_2 model were greater than those of the Clinical model (0.893 vs. 0.690, 84.4% vs. 71.9%, 88.9% vs. 66.7%, 84.6% vs. 64.7%), and the sensitivity and NPV were equivalent between the two models (78.6% vs. 78.6%, 84.2% vs. 80.0%), In the validation cohort (Table 4). Ultrasound images are presented in Fig. 5.

Table 6 DeLong testing between ultrasomics models and clinical model
Full size table
Fig. 4
figure 4

Diagnostic performance of ultrasomics models and clinical model. (A) The ROC curves of three models in training cohort. (B) The ROC curves of three models in validation cohort. Combined_1: the integration of B-mode ultrasound and contrast-enhanced ultrasound; Combined_2: the integration of B-mode ultrasound: contrast-enhanced ultrasound: and deep learning; AUC: area under the curve

Full size image
Fig. 5
figure 5

Region of Interest (ROI) for rectal cancer tumors. (A-F) A patient who had a good response after neoadjuvant chemoradiotherapy. (G-L) A patient who had a poor response after neoadjuvant chemoradiotherapy. A, G Raw images of tumors from B-mode ultrasound; D, J Raw images of tumors from contrast-enhanced ultrasound. B, E, H, K tracing tumors’ ROI (yellow curve represents the tracing pathway); C, F, I, L Confirming tumors’ ROI (green area)

Full size image

Discussion

In this research, we utilized the open-source software PyRadiomics and the pre-trained ResNet101 model to extract features based on BUS and CEUS images. Reduced feature’s dimension gradually and built four ultrasomics models with MLP classifiers. Selected the optimal model through DeLong testing algorithm and compared it with the Clinical model. The Combined_2 model demonstrated the best diagnostic performance among the models in the training group. Produced a better diagnostic performance in assessing good response compared to the Clinical model (AUC, 0.893[95%CI: 0.780-1.000] vs. 0.690[95%CI: 0.509–0.872]), with greater in accuracy (84.4% vs. 71.9%), specificity (88.9% vs. 66.7%) and PPV (84.6% vs. 64.7%), had no significant difference in sensitivity (78.6% vs. 78.6%) and NPV (84.2% vs. 80.0%). The advantage of this study is that using baseline data before nCRT to construct models to predict the efficacy after nCRT, which will guide clinical surgical decision-making and personalized treatment for patients.

In recent years, there had been frequent reports on the prediction of nCRT efficacy in patients with LARC based on radiomics, mostly based on single or multiple MRI sequences [15, 21,22,23,24,25]. It is worldwide accepted that MRI is the gold standard and patients require MRI after completion of CRT not only to assess response but also resectability. It has also been proven effective and robust in assessing residual disease and guiding patients to choose a watch-and-wait treatment strategy, and predicting survival as correlates with histopathology findings. However, there are very few reports on the prediction of nCRT in patients with LARC based on ultrasomics. This may be related to the difficulty in achieving the satisfactory quality of ultrasound images: firstly, the acquisition of each ultrasound image needs to be manually completed by the examining physician; Secondly, optimal imaging parameters vary for each patient; Finally, the patient’s own conditions and cooperation also influence the quality of the images. Ultrasound examination also has certain advantages, such as it is radiation-free and imaging time is significantly shorter than MRI and so on. This study constructs predictive models based on multimodal ultrasound images, and the results show that the transfer learning models based on BUS and CEUS have high diagnostic efficiency for identifying LARC patients with good response to nCRT. This may provide additional value for future multimodal radio studies in rectal cancer. Interestingly, a recent study reported the prediction of radiotherapy response in patients with rectal cancer based on transrectal ultrasound, which found that ultrasomics scores could serve as a biomarker to predict the pathological characteristics of rectal cancer [26]. Unfortunately, this study did not use DeLong test to compare the diagnostic performance between models, and also had a small sample size of only 43 cases. Our study had a relatively larger sample size than this study (106 vs. 43).

Previous studies had indicated that various clinical factors could affect the pathological reaction of nCRT in patients with LARC, including the CEA and CA199 level, and clinical staging [27, 28]. In our investigation, both univariate and multivariate logistic regression analyses show that age (P < 0.01) and clinical stage (P < 0.01) can be as an independent predictor of good response after nCRT in patients with LARC. It demonstrates that patients over 60 years of age and stage IV clinical staging, have a poorer response to nCRT, which is consistent with the findings of previous research. The AUC value of the clinical model was 0.690 in our study, which was consistent with the outcomes of previous research [29]. However, the diagnostic performance of the clinical model was significantly lower than the ultrasomics model in our study (AUC, 0.690 vs. 0.893, P < 0.01).

Comparing the prediction performance of four ultrasomics models through DeLong testing, which proves that the Combined_1 model shows greater than the BUS model in the validation cohort (P < 0.05) and the CEUS model in the training set (P < 0.05). Interestingly, although the AUC value of the Combined_2 model is superior than the Combined_1 model, its overall performance has no significant advantages regardless of whether it is the training group (AUC, [0.844 vs. 0.833, P = 0.774]) or the validation group (AUC, [0.893 vs. 0.829, P = 0.140]). This may be related to the pre-trained ResNet101 model. If we use a deep learning model that trained with our own images for transfer learning, its performance may be significantly improved, which is what we wanted to achieve in the future.

Although the results of this study are promising, there are several constraints: firstly, this study is a single-center retrospective study, which may result in selection bias and lack of generalizability. Prospective studies, multi- center data integration and external validation are essential in the future. Secondly, although we have adopted “ten-fold cross validation” to improve the model’s stability, the limited sample size still has an effect on it. Expanding the sample size to train the model is expected to improve its stability and increase clinical applicability in the future. Thirdly, we only use the pre-nCRT data, we will combine post-nCRT data to train the model in the future, which is expected to improve its predictive performance. Fourthly, we only use the pre-trained model to extract deep learning features. It is necessary to use deep transfer learning model to extract features to train the model in the future. Finally, we only use ultrasound images, and in the future, integrating multimodal data such as CT and MRI images to train the model is expected to improve its predictive performance.

Conclusion

In conclusion, the ultrasomics models show higher diagnostic performance than the clinical model to predict good response in patients with LARC after nCRT. This study indicates that ultrasomics scores have the prospective to become a non-invasive radiomics biomarker to predict efficacy in patients with LARC after nCRT in the future. It will provide valuable information for surgical decision-making by clinicians and personalized treatment for patients.

Data availability

All data generated or analysed during this study are included in this published article.

Abbreviations

AUC:

Area under the curve;

BUS:

B-mode ultrasound;

CEUS:

Contrast-enhanced ultrasound;

CI:

Confidence interval;

LARC:

Locally advanced rectal cancer;

LASSO:

Least absolute shrinkage and selection operator;

MLP:

Multilayer perceptron neural network;

nCRT:

Neoadjuvant chemoradiotherapy;

NPV:

Negative predictive value

pCR:

Pathological complete response;

PPV:

Positive predictive value;

ROI:

Region of interest;

TRG:

Tumor regression grade;

References

  1. Benson AB, Venook AP, Al-Hawary MM, Azad N, Chen YJ, Ciombor KK, Cohen S, Cooper HS, Deming D, Garrido-Laguna I, et al. Rectal cancer, version 2.2022, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw. 2022;20(10):1139–67.

    Article  CAS  PubMed  Google Scholar 

  2. Benson AB, Venook AP, Al-Hawary MM, Cederquist L, Chen YJ, Ciombor KK, Cohen S, Cooper HS, Deming D, Engstrom PF, et al. Rectal cancer, version 2.2018, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw. 2018;16(7):874–901.

    Article  PubMed  PubMed Central  Google Scholar 

  3. Maas M, Nelemans PJ, Valentini V, Das P, Rodel C, Kuo LJ, Calvo FA, Garcia-Aguilar J, Glynne-Jones R, Haustermans K, et al. Long-term outcome in patients with a pathological complete response after chemoradiation for rectal cancer: a pooled analysis of individual patient data. Lancet Oncol. 2010;11(9):835–44.

  4. Wang G, Zheng Z, Chen J, Ye J, Tang Z, Fang Y, Yao K, Zeng Q, Yang Y, Tang H, et al. A nomogram for predicting good response after neoadjuvant chemoradiotherapy for locally advanced rectal cancer: a retrospective, double-center, cohort study. Int J Colorectal Dis. 2022;37(10):2157–66.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Jiang H, Guo W, Yu Z, Lin X, Zhang M, Jiang H, Zhang H, Sun Z, Li J, Yu Y, et al. A comprehensive prediction model based on MRI radiomics and clinical factors to predict tumor response after neoadjuvant chemoradiotherapy in rectal cancer. Acad Radiol. 2023;30(Suppl 1):S185–98.

    Article  PubMed  Google Scholar 

  6. Shin J, Seo N, Baek SE, Son NH, Lim JS, Kim NK, Koom WS, Kim S. MRI radiomics model predicts pathologic complete response of rectal cancer following chemoradiotherapy. Radiology. 2022;303(2):351–8.

    Article  PubMed  Google Scholar 

  7. Giannini V, Mazzetti S, Bertotto I, Chiarenza C, Cauda S, Delmastro E, Bracco C, Di Dia A, Leone F, Medico E, et al. Predicting locally advanced rectal cancer response to neoadjuvant therapy with (18)F-FDG PET and MRI radiomics features. Eur J Nucl Med Mol Imaging. 2019;46(4):878–88.

    Article  CAS  PubMed  Google Scholar 

  8. Bulens P, Couwenberg A, Intven M, Debucquoy A, Vandecaveye V, Van Cutsem E, D’Hoore A, Wolthuis A, Mukherjee P, Gevaert O, et al. Predicting the tumor response to chemoradiotherapy for rectal cancer: model development and external validation using MRI radiomics. Radiother Oncol. 2020;142:246–52.

    Article  CAS  PubMed  Google Scholar 

  9. Wu YQ, Gao RZ, Lin P, Wen R, Li HY, Mou MY, Chen FH, Huang F, Zhou WJ, Yang H, et al. An endorectal ultrasound-based radiomics signature for preoperative prediction of lymphovascular invasion of rectal cancer. BMC Med Imaging. 2022;22(1):84.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Jiang M, Li CL, Luo XM, Chuan ZR, Lv WZ, Li X, Cui XW, Dietrich CF. Ultrasound-based deep learning radiomics in the assessment of pathological complete response to neoadjuvant chemotherapy in locally advanced breast cancer. Eur J Cancer. 2021;147:95–105.

    Article  CAS  PubMed  Google Scholar 

  11. Gu J, Tong T, He C, Xu M, Yang X, Tian J, Jiang T, Wang K. Deep learning radiomics of ultrasonography can predict response to neoadjuvant chemotherapy in breast cancer at an early stage of treatment: a prospective study. Eur Radiol. 2022;32(3):2099–109.

    Article  PubMed  Google Scholar 

  12. DiCenzo D, Quiaoit K, Fatima K, Bhardwaj D, Sannachi L, Gangeh M, Sadeghi-Naini A, Dasgupta A, Kolios MC, Trudeau M, et al. Quantitative ultrasound radiomics in predicting response to neoadjuvant chemotherapy in patients with locally advanced breast cancer: results from multi-institutional study. Cancer Med. 2020;9(16):5798–806.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Mao B, Ma J, Duan S, Xia Y, Tao Y, Zhang L. Preoperative classification of primary and metastatic liver cancer via machine learning-based ultrasound radiomics. Eur Radiol. 2021;31(7):4576–86.

    Article  PubMed  Google Scholar 

  14. Zhang C, Liu D, Huang L, Zhao Y, Chen L, Guo Y. Classification of thyroid nodules by using deep learning radiomics based on ultrasound dynamic video. J Ultrasound Med. 2022;41(12):2993–3002.

    Article  PubMed  Google Scholar 

  15. Shaish H, Aukerman A, Vanguri R, Spinelli A, Armenta P, Jambawalikar S, Makkar J, Bentley-Hibbert S, Del Portillo A, Kiran R, et al. Radiomics of MRI for pretreatment prediction of pathologic complete response, tumor regression grade, and neoadjuvant rectal score in patients with locally advanced rectal cancer undergoing neoadjuvant chemoradiation: an international multicenter study. Eur Radiol. 2020;30(11):6263–73.

    Article  CAS  PubMed  Google Scholar 

  16. Song C, Chung JH, Kang SB, Kim DW, Oh HK, Lee HS, Kim JW, Lee KW, Kim JH, Kim JS. Impact of tumor regression grade as a major prognostic factor in locally advanced rectal cancer after neoadjuvant chemoradiotherapy: a proposal for a modified staging system. Cancers (Basel). 2018;10(9).

  17. Ji GW, Zhu FP, Xu Q, Wang K, Wu MY, Tang WW, Li XC, Wang XH. Radiomic features at contrast-enhanced CT predict recurrence in early stage hepatocellular carcinoma: a multi-institutional study. Radiology. 2020;294(3):568–79.

    Article  PubMed  Google Scholar 

  18. Lin P, Lin YQ, Gao RZ, Wan WJ, He Y, Yang H. Integrative radiomics and transcriptomics analyses reveal subtype characterization of non-small cell lung cancer. Eur Radiol. 2023;33(9):6414–25.

    Article  CAS  PubMed  Google Scholar 

  19. Huang Y, Zhu T, Zhang X, Li W, Zheng X, Cheng M, Ji F, Zhang L, Yang C, Wu Z, et al. Longitudinal MRI-based fusion novel model predicts pathological complete response in breast cancer treated with neoadjuvant chemotherapy: a multicenter, retrospective study. EClinicalMedicine. 2023;58:101899.

    Article  PubMed  PubMed Central  Google Scholar 

  20. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–45.

    Article  CAS  PubMed  Google Scholar 

  21. Horvat N, Veeraraghavan H, Khan M, Blazic I, Zheng J, Capanu M, Sala E, Garcia-Aguilar J, Gollub MJ, Petkovska I. MR imaging of rectal cancer: radiomics analysis to assess treatment response after neoadjuvant therapy. Radiology. 2018;287(3):833–43.

    Article  PubMed  Google Scholar 

  22. Yang L, Qiu M, Xia C, Li Z, Wang Z, Zhou X, Wu B. Value of high-resolution DWi in combination with texture analysis for the evaluation of tumor response after preoperative chemoradiotherapy for locally advanced rectal cancer. AJR Am J Roentgenol. 2019;212(6):1279–86.

    Article  PubMed  Google Scholar 

  23. Aker M, Ganeshan B, Afaq A, Wan S, Groves AM, Arulampalam T. Magnetic resonance texture analysis in identifying complete pathological response to neoadjuvant treatment in locally advanced rectal cancer. Dis Colon Rectum. 2019;62(2):163–70.

    Article  PubMed  Google Scholar 

  24. Cui Y, Yang X, Shi Z, Yang Z, Du X, Zhao Z, Cheng X. Radiomics analysis of multiparametric MRI for prediction of pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Eur Radiol. 2019;29(3):1211–20.

    Article  PubMed  Google Scholar 

  25. Liu Z, Zhang XY, Shi YJ, Wang L, Zhu HT, Tang Z, Wang S, Li XT, Tian J, Sun YS. Radiomics analysis for evaluation of pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. Clin Cancer Res. 2017;23(23):7253–62.

    Article  CAS  PubMed  Google Scholar 

  26. Abbaspour S, Abdollahi H, Arabalibeik H, Barahman M, Arefpour AM, Fadavi P, Ay M, Mahdavi SR. Endorectal ultrasound radiomics in locally advanced rectal cancer patients: despeckling and radiotherapy response prediction using machine learning. Abdom Radiol (NY). 2022;47(11):3645–59.

    Article  PubMed  Google Scholar 

  27. Song M, Li S, Wang H, Hu K, Wang F, Teng H, Wang Z, Liu J, Jia AY, Cai Y, et al. MRI radiomics independent of clinical baseline characteristics and neoadjuvant treatment modalities predicts response to neoadjuvant therapy in rectal cancer. Br J Cancer. 2022;127(2):249–57.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  28. Zheng Z, Wang X, Lu X, Huang Y, Chi P. Prognostic significance of carcinoembryonic antigen combined with carbohydrate antigen 19– 9 following neoadjuvant chemoradiotherapy in patients with locally advanced rectal cancer. Colorectal Dis. 2021;23(9):2320–30.

    Article  CAS  PubMed  Google Scholar 

  29. Delli Pizzi A, Chiarelli AM, Chiacchiaretta P, d’Annibale M, Croce P, Rosa C, Mastrodicasa D, Trebeschi S, Lambregts DMJ, Caposiena D, et al. MRI-based clinical-radiomics model predicts tumor response before treatment in locally advanced rectal cancer. Sci Rep. 2021;11(1):5379.

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

We thank the Key Laboratory of Ultrasonic Molecular Imaging and Artificial Intelligence, Guangxi Zhuang Autonomous Region Engineering Research Center for Artificial Intelligence Analysis of Multimodal Tumor Images, and Guangxi Key Laboratory of Early Prevention and Treatment for Regional High Frequency Tumor/Key Laboratory of Early Prevention and Treatment for Regional High Frequency Tumor (Guangxi Medical University), Ministry of Education for their support of this study.

Funding

This study funded by Selfraised Scientific Research Funds of Ministry of Health of Guangxi Province (No. Z20201070).

Author information

Author notes

  1. Qiong Qin and Xiangyu Gan are first authors.

Authors and Affiliations

  1. Department of Medical Ultrasound, The First Affiliated Hospital of Guangxi Medical University, No.6 Shuangyong Road, Nanning, Guangxi Zhuang Autonomous Region, 530021, China

    Qiong Qin, Xiangyu Gan, Peng Lin, Jingshu Pang, Ruizhi Gao, Rong Wen, Dun Liu, Quanquan Tang, Changwen Liu, Yun He, Hong Yang & Yuquan Wu

Authors
  1. Qiong Qin
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Xiangyu Gan
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Peng Lin
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jingshu Pang
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Ruizhi Gao
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Rong Wen
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Dun Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Quanquan Tang
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Changwen Liu
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Yun He
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Hong Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Yuquan Wu
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

QQ, GXY, LP and PJS contributed to the manuscript preparation and data analysis. GRZ, WR and LD prepared the Figures. TQQ and LCW contributed to the data collection. HY, YH and WYQ contributed to the founding of the entire study and manuscript editing. All authors provided final approval of the version submitted for publication.

Corresponding authors

Correspondence to Hong Yang or Yuquan Wu.

Ethics declarations

Ethics approval and consent to participate

This retrospective study was conducted with approval from the Medical Ethics Committee of First Affiliated Hospital of Guangxi Medical University (No.2023-E276-01). The requirement of informed consent was waived by the Ethics Committee of First Affiliated Hospital of Guangxi Medical University owing to the retrospective nature of the study. This study was performed in accordance with the ethical standards of the institutional and national research committees as well as the Helsinki Declaration.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Qin, Q., Gan, X., Lin, P. et al. Development and validation of a multi-modal ultrasomics model to predict response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer. BMC Med Imaging 24, 65 (2024). https://doi.org/10.1186/s12880-024-01237-0

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12880-024-01237-0

Keywords

BMC Medical Imaging

ISSN: 1471-2342

Contact us