Can pre-treatment MRI and CT radiomics improve prediction of complete pathological response in locally advanced rectal cancer following neoadjuvant treatment?

doi:10.21203/rs.3.rs-3756695/v1

Download PDF

Research Article

Can pre-treatment MRI and CT radiomics improve prediction of complete pathological response in locally advanced rectal cancer following neoadjuvant treatment?

https://doi.org/10.21203/rs.3.rs-3756695/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Objective(s):

The treatment response to neoadjuvant chemoradiation (nCRT) differs largely in individuals treated for rectal cancer. In this study we investigated the role of radiomics to predict the pathological response in locally advanced rectal cancers at different treatment time points. 1) prior to start of any treatment using baseline T2-weighted MRI (T2W-MR) 2) at the start of radiation treatment using planning CT.

Methods

Patients who received nCRT followed by surgery between June 2017 to December 2019 were included in the study. Based on the histopathological tumour response grading (TRG) criteria, 58 patients with TRG 1 were classified as complete responders (pCR) and the rest as incomplete responders (IR). The gross tumour volume was defined by the radiation oncologists. Following resampling, 100 and 103 Pyradiomic features were extracted from T2W-MR and planning CT images, respectively. Synthetic minority oversampling technique (SMOTE) was applied to correct for class imbalance. Four Machine Learning classifiers were used to build clinical, radiomics and merged models. Model performance was evaluated on a held-out validation dataset following repeated stratified 3-fold cross validation using area-under-the-receiver-operator-characteristic curves (AUC) with bootstrap 95% confidence intervals.

Results

150 patients were included. Clinical models performed better (AUC = 0.68) than the radiomics models (AUC = 0.62). Overall, the clinical + T2W-MR merged model performed the best (AUC = 0.72) in predicting the pathological response prior to therapy. Clinical + Planning CT merged models could only achieve highest AUC = 0.66.

Conclusion

Combining clinical and radiomics from baseline T2W-MR improves the prediction of pathological response in rectal cancer. Validation in larger cohorts is warranted before they can guide clinical decisions; especially in watch and wait strategies.

Oncology

Artificial Intelligence and Machine Learning

Rectal cancer

Neoadjuvant chemoradiation

Pathological complete response

Radiomics

Machine learning

Colorectal cancer (CRC) is the third most diagnosed cancer and is the fourth cause of death by cancer in the world. According to the predictions, by 2030, CRC will be responsible for 2.2 million new cancer cases and 1.1 million cancer deaths (1). Rectal cancer (RC) corresponds approximately to one third of the CRC cases (2, 3) and presents unique challenges in terms of staging and treatment.

Based on the results of the PRODIGE 23 (4) and RAPIDO trials (5) the recommended treatment for locally advanced rectal cancer (LARC) is neoadjuvant chemoradiotherapy (nCRT) with or without neoadjuvant chemotherapy followed by radical resection and adjuvant chemotherapy when lymph nodes are involved. After nCRT about 15% of patients have a Pathological Complete Response (pCR), defined by the absence of viable tumor cells (ypT0N0M0) in the surgical resection specimen and these patients present a better 5-year disease free survival (DFS). However, there is an unmet need to be able to predict the clinical or pathological tumour response to nCRT prior to initiation of treatment. Being able to predict patients who are unlikely to achieve a pCR can help identify high-risk patients who may need more aggressive treatment to improve their outcomes. There may be potential in escalating therapy in predicted non-responders and offer local excision or wait-and-watch strategies in patients more likely to have a pCR (5–10). Thus, being able to predict the outcome of nCRT before initiating treatment can optimise choice of treatment and can avoid overtreatment or undertreatment.

Predictive models have been developed to estimate the likelihood of a pCR based on various patient and tumour characteristics, such as age, gender, tumour stage, and other clinical features (11–16). Radiomics based models have also been explored to help predict pathological complete response in patients undergoing treatment for locally advanced rectal cancer (17, 18). Radiomics involves the extraction of large amounts of quantitative data from medical imaging, such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Positron Emission Tomography (PET) and Ultrasound (US) (19). These imaging features are textural, morphological, or functional and are often used to create a "radiomic signature" that can be used for a variety of purposes, such as predicting a patient's response to treatment or identifying early signs of disease (20–23).

The aim of this study was to investigate the role of radiomics in predicting the pathological response in patients treated for locally advanced rectal cancer at two different time points namely, 1) prior to start of any treatment using baseline T2-weighted MRI images 2) at the start of radiation treatment using planning CT images. We also studied the prediction performance of the models when radiomic features are combined with clinical predictors.

Patient cohort

This is a single institutional, retrospective study involving patients with locally advanced rectal cancer who received concurrent chemoradiation followed by total mesorectal excision (TME) between January 2017 and December 2019 at our institution. The treatment details are mentioned in Appendix A1.

Only patients who received neoadjuvant chemotherapy, had baseline MR images with T2-weighted sequence, had radiation treatment planning CT, underwent surgery at CMC, and had complete histopathological report were included in this study.

Poor image quality, unavailability of CT / MR acquired at our institution, and metastatic disease were reasons for patient exclusion. The primary endpoint was the prediction of pCR to neoadjuvant chemoradiation, specified as the absence of disease on the surgical specimen. The study was approved with waiver of informed patient consent by the institution review board and was conducted in accordance with the Helsinki declaration guidelines and regulations.

Image acquisition

The patients were imaged in headfirst supine position for both MRI and planning CT. T2-weighted MRI (T2W-MRI) were acquired as part of diagnosis at baseline on 1.5 T (Achieva, Philips Healthcare) or 3.0 T (Avanto fit, Siemens Healthineers). Planning CT scans were acquired in patients after chemotherapy for those who received Total Neoadjuvant therapy (TNT), and upfront in those who received chemoradiation, respectively. For majority of the patients contrast enhanced CT was acquired on Siemens SOMATOM definition AS or Siemens Biograph 6. Further imaging characteristics are summarized in Supplementary Table A1 and A2.

Tumour Delineation

The primary gross tumour volume (GTVp) was manually outlined separately on the CT and MRI by trained radiation oncologists (J.K.R, A.S, N.O.J) with more than 5 years of experience in colorectal cancers on axial T2W-MRI, were blinded to the histopathological outcome. The tumours were defined using Eclipse Treatment Planning System (v16.1) to include the demonstrable extent of the gross primary tumour on the baseline T2-Weighted MRI and on the radiotherapy planning CT. Given the timepoint, the information from the clinical examination, colonoscopy and available T2-weighted MRI was used to influence the extent of the tumour boundary. All segmentations were verified by the senior radiation oncologist.

Pathological Assessment

Following total mesorectal excision (TME), each patient’s pathological response was assessed according to the modified Ryan’s tumor regression grade (TRG). Patients with TRG0 were considered to have pCR while TRG1, TRG2, and TRG3 were considered poor responders and collectively classified as incomplete responders (IR) for this study.

Radiomics Feature extraction

Pyradiomics (v3.0.1) along with the wrapper software Ontology Guided Radiomics Analysis Workflow (O-RAW) was used to extract the radiomics features separately for T2-Weighted MRI and planning CT. All radiomics features were extracted on the original image. Prior to extraction, the CT and MRI were resampled to isotropic voxel size of 3 mm and 2 mm, respectively and the intensity values were discretised to a fixed bin width of 25 and 32, respectively. For each ROI, first-order, shape, GrayLevel Co-occurrence matrix (GLCM), Gray level run length matrix (GLRLM), Gray level size zone matrix (GLSZM), Gray level dependence matrix (GLDM) and Neighbourhood Gray Tone distance matrix (NGTDM) features were calculated. The parameter files used to extract the MRI and CT features can be found in the appendix. A total of 100 and 103 radiomic features were obtained for the MR and planning CT images, respectively. The parameter files used for the CT and T2W-MR feature extraction is shared in the supplementary section. (Appendix 2 and 3)

Radiomics Feature selection: Three different feature selection methods were considered for this study and implemented in Python (v3.7) namely Boruta_py (v0.3), with Random Forest as its estimator, SelectKbest and LASSO. The feature selection was developed as a three-step approach. The first step involved selecting the features from one of the three selection methods based on the predictive value of the pCR status. The features that appeared more than 250 times in the 1000 iterations were retained. As a second step, based on their inter-correlation using the Spearman’s rank correlation coefficient (ρ): when two features were highly correlated (ρ > 0.8), only the feature with the highest AUC was retained. In the last step, features that correlated highly with the tumour volume feature were removed. The features that were selected at the end of the three-step process were fed into the machine learning classifiers. Also, a combined feature set was created using the features selected by the above three methods and considered for modelling.

Experimental Set-up and Statistical Analysis:

The complete workflow can be seen from Fig. 1. Four classifiers namely Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF) and Multi-layer Perceptron (MLP) were built with default scikit-learn (v1.0.2) library parameters and repeated stratified 3-fold cross-validation method with 3 repetitions. For each classifier, the ML model hyperparameters were tuned using grid search hyperparameter optimization algorithm. Before building each model, data imbalance was corrected using Synthetic oversampling of minority class technique (SMOTE). The data was stratified into training and test groups while maintaining the ratio of IR to pCR to be the similar in the two groups. Six models were built using clinical and radiomics features from T2W-MRI and Planning CT. For robustness optimization, internal validation was performed using bootstrapping that had n = 1000 replications. Each model’s performance was evaluated using the Area Under the Receiver Operation Characteristic (ROC) Curve (AUC). The classification characteristics of the AUC namely Accuracy, Specificity, Sensitivity averaged was also evaluated. Each model was then evaluated across the 1000 stratified bootstraps on the testing set across for AUC and 95% confidence intervals. The model parameters are shown in the Supplementary Table A3, and the final selected model parameters used for the validation data is described in the Supplementary Tables A4 – A8.

Clinical data was analysed with either Chi-squared test or Fisher's exact test in case of categorical variables. Mann–Whitney U test was used for numerical variables and P-value less than 0.05 was considered statistically significant. The statistical analyses were performed with Python SciPy (v1.7.3).

Patient characteristics

From January 2017 to December 2019, 336 patients who underwent neoadjuvant treatment for LARC between for locally advanced rectal carcinoma followed by TME surgery were screened and the flowchart is shown in Fig. 2. 150 patients were included in the study and their clinical baseline and post-operative characteristics are shown in Table 1. Among 150 patients, 80 received Chemoradiation either with oral capecitabine or 5-Fluorouracil and 70 patients received total neoadjuvant therapy (TNT) (i.e) neoadjuvant chemotherapy followed by radiation with or without concurrent chemotherapy. All patients in this cohort underwent surgery. Majority of the patients were male (65%), had T3 (69%) and T4 (25%) stage disease and 89% of the patients had lymph node involvement. Majority (60%) had moderately differentiated adenocarcinoma and 17% did not have differentiation available. CEA levels were, <=5 ng/mL in 52% of the patients and > 15 ng/mL in 19% of patients. 70% of the patients had involved MR assessed circumferential resection margin (mrCRM) and 37% had extramural venous invasion (EMVI); signet ring cells were seen in 11% and pCR was significantly more common among younger patients with median age of 42 years vs. 50 years) (p = 0.04), moderately differentiated disease (49%) compared to pCR group (69%) p = 0.02); no signet ring cell occurrence (p < 0.0001). All patients’ characteristics are available in Table 1. Totally 58 patients achieved pCR following neoadjuvant chemoradiation.

Table 1

Patient characteristics
Baseline
Variable	Overall population n (%)	pCR n (%)	IR n (%)	P value*
Age, years				0.009^#
Median (IQR)	47(19–79)	42.5(19–69)	50 (20–79)
Sex				0.863^$
Male	97 (64.7%)	37 (63.8%)	60 (65.2%)
Female	53 (35.3%)	21 (36.2%)	32 (34.8%)
T stage				0.218^$
2	10 (6.7%)	5 (8.6%)	5 (5.4%)
3	103 (68.7%)	35 (60.3%)	68 (73.9%)
4	37 (24.7%)	18 (31.0%)	19 (20.7%)
N stage				0.173^$
0	16 (10.7%)	8 (13.8%)	8 (8.7%)
1	50 (33.3%)	23 (39.7%)	27 (29.3%)
2	84 (56.0%)	27 (46.6%)	57 (62.0%)
Tumour Differentiation				0.016^$
Gr 1	7 (4.7%)	4 (6.9%)	3 (3.3%)
Gr 2	90 (60.0%)	27 (46.6%)	63 (68.5%)
Gr 3	27 (18.0%)	17 (29.3%)	10 (10.9%)
Gr 4	25 (16.7%)	9 (15.5%)	16 (17.4%)
Gr 5	1 (0.7%)	1 (1.7%)	0 (0%)
Signet-ring cell Occurrence				< 0.001*
Yes	17 (11.3%)	14 (24.1%)	3 (3.3%)
No	133 (88.7%)	44 (75.9%)	89 (96.7%)
CEA ng/mL				0.091^$
<=5	78 (52.0%)	37 (63.8%)	41 (44.6%)
>5 < = 15	44 (29.3%)	13 (22.4%)	31 (33.7%)
>15	28 (18.7%)	8 (13.8%)	20 (21.7%)
Nodal disease				0.787*
Yes	134 (89.3%)	51 (87.9%)	83 (90.2%)
No	16 (10.7%)	7 (12.1%)	9 (9.8%)
Biopsy mucin levels				0.14^$
Range 1	130 (86.7%)	47 (81.0%)	83 (90.2%)
Range 2	19 (12.7%)	11 (19.0%)	8 (8.7%)
Range 3	1 (0.7%)	0 (0%)	1 (1.1%)
EMVI				0.603*
Yes	55 (36.7%)	23 (39.7%)	32 (34.8%)
No	95 (63.3%)	35 (60.3%)	60 (65.2%)
mrCRM				0.838^$
Involved	106 (70.7%)	42 (72.4%)	64 (69.6%)
Not involved	37 (24.7%)	14 (24.1%)	23 (25.0%)
Threatened	7 (4.7%)	2 (3.4%)	5 (5.4%)
Post-Operative
Variable	Overall population n (%)	pCR n (%)	IR n (%)	P value*
Ryan’s Grading				< 0.001^$
TRG 1	58 (38.7%)	58 (100.0%)	0 (0%)
TRG 2	22 (14.7%)	0 (0%)	22 (23.9%)
TRG 3	46 (30.7%)	0 (0%)	46 (50.0%)
TRG 4	24 (16.0%)	0 (0%)	24 (26.1%)
Pathological T stage				< 0.001^$
0	58 (38.7%)	58 (100.0%)	0 (0%)
1	7 (4.7%)	0 (0%)	7 (7.6%)
2	31 (20.7%)	0 (0%)	31 (33.7%)
3	51 (34.0%)	0 (0%)	51 (55.4%)
4	3 (2.0%)	0 (0%)	3 (3.3%)
Pathological N stage				0.001^$
0	122 (81.3%)	56 (96.6%)	66 (71.7%)
1	24 (16.0%)	2 (3.4%)	22 (23.9%)
2	4 (2.7%)	0 (0%)	4 (4.3%)
Pathological M stage				1.000*
0	149 (99.3%)	58 (100.0%)	91 (98.9%)
1	1 (0.7%)	0 (0%)	1 (1.1%)
pCRM				0.381^$
pCRM(+)	3 (2.0%)	0 (0%)	3 (3.3%)
pCRM (-)	147 (98.0%)	58 (100.0%)	89 (96.7%)
Distal margin involvement				1.000*
Yes	0 (0%)	0 (0%)	0 (0%)
No	150 (100.0%)	58 (100.0%)	92 (100.0%)
Proximal margin involvement				1.000*
Yes	1 (0.7%)	0 (0%)	1 (1.1%)
No	149 (99.3%)	58 (100.0%)	91 (98.9%)
Note: CEA, carcinoembryonic antigen; EMVI Extramural venous invasion; mrCRM, MRI-assessed circumferential margin; pCRM, pathological circumferential margin
* Fisher exact, ^$chi-square, ^#Mann- Whitney

The clinical predictors of pCR in this study cohort were age, tumour biopsy differentiation and presence of signet-ring cell. The clinical model however; included 11 baseline clinical features that were reported as predictors for pCR and the model achieved an AUC of 0.841, 0.69 and 0.68 in the training, testing and validation data, respectively.

After feature selection only 4 from T2W-MR sequence remained, of which 2 were from Gray level Size zone matrix; one Gray Level Difference matrix texture feature and one was Kurtosis and 2 from planning CT (Complexity and Elongation). The difference in the feature values between patients who has pCR and IR is also shown in Table 2.

Table 2

Selected feature statistics for T2W-MR and Planning CT radiomics
Modality	Radiomic feature	Median feature value (pCR)	Median feature value (IR)	P value
T2W-MR	GLSZM Large Area Low Gray Level Emphasis	0.747	1.573	0.013
	GLDM Gray Level Nonuniformity	124.323	163.303	0.006
	GLSZM Large Area Emphasis	14.841	21.173	0.027
	Kurtosis	3.040	3.547	0.080
Planning CT	NGTDM Complexity	51.214	35.100	0.491
Planning CT	Elongation	0.482	0.503	0.194

The model with the best classifier is shown in Table 3 and the corresponding ROC are shown in Fig. 2. The clinical + planning CT merged model performed the best in the training (AUC = 0.98) and testing (AUC = 0.81). However, the AUC values were most efficient in the Clinical + T2W-MR merged model (AUC = 0.72). Further information on selected features and its classification performance for each modelling task can be found in supplementary Tables (A3 – A7). Table 4 shows the classification of the Ryan’s tumour regression grade. In that the clinical model was the best performing model with AUC of 0.76, 0.68 and 0.58 in the training, test and validation. The modelling parameters and hyperparameters with the classification results can be found in supplementary Table A8.

Table 3

Prediction model results summarizing the best performing classifier and metrics for each modelling task
Model	Best classifier	Train set	Test set	Validation set
Clinical	RF	0.841 (0.840–0.842)	0.689 (0.687–0.691)	0.676 (0.674–0.678)
T2W-MR Radiomics	RF	0.778 (0.777–0.779)	0.709 (0.707–0.711)	0.621 (0.620–0.622)
Planning CT Radiomics	LR	0.602 (0.600-0.603)	0.555 (0.553–0.557)	0.618 (0.616–0.620)
Clinical + T2W-MR Radiomics	RF	0.797 (0.796–0.798)	0.706 (0.705–0.708)	0.718 (0.716–0.719)
Clinical + Planning CT Radiomics	SVM	0.984 (0.984–0.985)	0.813 (0.812–0.815)	0.658 (0.656–0.659)
Clinical + T2W-MR + Planning CT Radiomics	LR	0.791 (0.791–0.792)	0.685 (0.683–0.687)	0.664 (0.663–0.666)

Note: Metrics shown with mean AUC and 95% Confidence Intervals

Table 4

Ryan’s TRG classification results summarizing the best performing classifier and metrics for each modelling task
Model	Best classifier	Train set	Test set	Validation set
Clinical	MLP	0.762 (0.762–0.763)	0.684 (0.684–0.685)	0.581 (0.580–0.581)
T2W-MR Radiomics	SVM	0.991 (0.991–0.991)	0.810 (0.810–0.811)	0.567 (0.566–0.567)
Planning CT Radiomics	RF	0.650 (0.649–0.650)	0.619 (0.619–0.620)	0.524 (0.523–0.525)
Note: Metrics shown with mean AUC and 95% Confidence Intervals

In this study, we evaluated several machine learning models for early assessment of pathological response to neoadjuvant chemoradiation in patients treated for locally advanced rectal cancer. In this study population from a single centre, we found that merged model using MR radiomics and clinical features performed better than the individual clinical and radiomics models.

Neoadjuvant chemoradiation followed by surgery is the standard of care in locally advanced Rectal cancers (1). However, there is a shift in the treatment strategy towards total neoadjuvant therapy (TNT) and most of the trials which were designed with the intention of improving distant control, also demonstrated improvement in pathological complete response rates (2). Literature suggests that patients who have a pathological complete response (pCR) have a better prognosis when compared to those who do not have a complete response (24). Patients who achieve pCR have better local control, better survival, and less distant failure (3) and demonstrated improved rectum preservation leading to better quality of life in these patients. If these patients could be predicted upfront, then they could be offered organ preservation strategies especially in low-risk rectal cancers (25). And if the patients with imaging features portending poor response and early systemic dissemination can be predicted, then their treatment could be intensified before surgery (26).

Till date clinical evaluation is the most widely used assessment tool for identifying pCR following nCRT. Certain clinical features such as tumor stage, nodal stage, histology and pretreatment Carcinoembryonic Antigen (CEA) have been reported to predict pathological complete response in rectal cancers (4). MERCURY trial demonstrated the importance of pretreatment high resolution MRI in assessing the local extensions which had impact on the survival (27) and outcomes.

Pretreatment Radiomics have shown some promising directions towards prediction of outcomes and survival in rectal cancer (20, 21, 24). In this study, we explored Radiomics from pre-treatment T2W-MRI and post chemotherapy Planning CT; which has not really been evaluated. While the performance of the T2W-MRI and Planning CT models were quite similar (AUC = 0.62) in predicting pCR, the feature values themselves were different in patients who went on to have pCR compared to the IR group. However, the CT features did not show significant difference between the pCR and IR groups. The merged model with T2W radiomics and clinical features performed the best and was able to discriminate between patients who achieved pCR and IR with an area under the curve of 0.80 and 0.72 in the training and the validation cohorts, respectively. The combined model also showed better performance than models built with individual features alone (Fig. 3). However, it should be noted that the low predictive values for most of the models must be keeping in mind the low prevalence of rectal cancer and even lower rates of pCR in any given cohort. So, it is quite difficult for single institutional data to account for the variabilities that negatively impacts the generalizability of models to real-life data such as the variations seen across populations, image quality and size of the dataset. Restaging MRI has shown promise in earlier studies for predicting complete response (28) and it would be interesting to evaluate if combining radiomics based prediction using these images can further improve the accuracy for predicting complete response and follow-up in wait and watch strategy. Given the growing number of studies on deep learning based oncological prognostication and the ability of these models to handle the heterogeneity in the data better compared to machine learning (29), it would be worth exploring their utility for future studies. It would also be interesting to see if with federated architecture (30) where we can have more access to more data without having to worry much about data leaving the participating institutions; can help us improve the predictive performance of the models for future clinical use.

Our study has a few limitations. This is a single institute retrospective study. The models were trained and validated in a relatively small sample of patients. There were variations in the image acquisition parameters for CT and MRI. The datasets included have both contrast and non-contrast enhanced CTs; with patients who underwent TNT having received contrast enhanced CT, while others got non-contrast enhanced CTs. However, we did not correct this since the impact of contrast agents on radiomics is not completely understood (31). The ideal situation would be to define the extent of the tumor on T2W high resolution MRI (32). Since all patients did not have high resolution MR images, we used the available T2W-MR images instead. The retrospective nature and smaller sample size of the study population may have limited the overall ML model performance. Large multicenter studies (17, 20, 33, 34) show that the ML model performs better with larger sample size for training and validation cohorts. Multi modal studies like (18) and RAPIDS, radio-pathomics based study (17) also show multimodal prediction models perform better than single-modality prediction models which was consistent with our results as the merged T2W MRI-clinical model achieved the best AUC. The performance of the Ryan’s grading classification model could not achieve an AUC > 0.58. (Table 4). This may also be attributed to a smaller number of patients in each grade which limits optimal training and validation of the models. Next, the model has not been validated with data from another institution to comment on the generalizability of the results. The correlation of pathological response and radiomics for rectal cancer has not been reported earlier in Indian patients. So future studies will address this limitation.

In summary, by combining information from clinical parameters and baseline T2W-MRI radiomics could predict pCR better than individual prediction models for rectal cancer. The Random Forest model had high predictive value. Before clinical use with a wait-and-watch strategy, the proposed models will require validation in larger clinical studies.

Acknowledgements:

The authors acknowledge the support provided by Ms. Jennifer Sneha for image retrieval.

Declarations of interest:

None

Funding:

This work was approved by CMC Institutional Review Board (IRB min no. 13855, 24.03.2021) and funded by CMC Major Research Grant

Data Availability

Research data are not available currently.

Feeney G, Sehgal R, Sheehan M, Hogan A, Regan M, Joyce M, et al. Neoadjuvant radiotherapy for rectal cancer management. World J Gastroenterol. 2019;25(33):4850–69.
Hoendervangers S, Burbach JPM, Lacle MM, Koopman M, van Grevenstein WMU, Intven MPW, et al. Pathological Complete Response Following Different Neoadjuvant Treatment Strategies for Locally Advanced Rectal Cancer: A Systematic Review and Meta-analysis. Ann Surg Oncol. 2020;27(11):4319–36.
Shin JK, Huh JW, Lee WY, Yun SH, Kim HC, Cho YB, et al. Clinical prediction model of pathological response following neoadjuvant chemoradiotherapy for rectal cancer. Sci Rep. 2022;12(1):7145.
Conroy T, Bosset JF, Etienne PL, Rio E, François É, Mesgouez-Nebout N, et al. Neoadjuvant chemotherapy with FOLFIRINOX and preoperative chemoradiotherapy for patients with locally advanced rectal cancer (UNICANCER-PRODIGE 23): a multicentre, randomised, open-label, phase 3 trial. Lancet Oncol. 2021;22(5):702–15.
Bahadoer RR, Dijkstra EA, van Etten B, Marijnen CAM, Putter H, Kranenbarg EMK, et al. Short-course radiotherapy followed by chemotherapy before total mesorectal excision (TME) versus preoperative chemoradiotherapy, TME, and optional adjuvant chemotherapy in locally advanced rectal cancer (RAPIDO): a randomised, open-label, phase 3 trial. Lancet Oncol. 2021;22(1):29–42.
López-Campos F, Martín-Martín M, Fornell-Pérez R, García-Pérez JC, Die-Trill J, Fuentes-Mateos R, et al. Watch and wait approach in rectal cancer: Current controversies and future directions. World J Gastroenterol. 2020;26(29):4218–39.
Smith JJ, Paty PB, Garcia-Aguilar J. Watch and Wait in Rectal Cancer or More Wait and See? JAMA Surg. 2020;155(7):657–8.
Wang Q xuan, Zhang R, Xiao W wei, Zhang S, Wei M biao, Li Y heng, et al. The watch-and-wait strategy versus surgical resection for rectal cancer patients with a clinical complete response after neoadjuvant chemoradiotherapy. Radiat Oncol. 2021;16(1):16.
Khalid A, Aloul Z, Chouhan H, Khalid A, Aloul Z, Chouhan H. Radical vs. Local Excision in Rectal Carcinoma T1N0M0: Recurrence and Mortality Rates. Cureus [Internet]. 2022 May 28 [cited 2023 Mar 22];14(5). Available from: https://www.cureus.com/articles/98142-radical-vs-local-excision-in-rectal-carcinoma-t1n0m0-recurrence-and-mortality-rates
Smits LJH, van Lieshout AS, Grüter AAJ, Horsthuis K, Tuynman JB. Multidisciplinary management of early rectal cancer – The role of surgical local excision in current and future clinical practice. Surg Oncol. 2022;40:101687.
Attia AM, Farrag A, Attia NM, Khalaf LM, Hassan HM, Ameen MG, et al. Signet ring cell component predicts the response to neoadjuvant chemoradiotherapy in rectal cancer. Long interim results of a single institution experience. Am J Cancer Res. 2022;12(3):1156–68.
Bates DDB, Homsi ME, Chang KJ, Lalwani N, Horvat N, Sheedy SP. MRI for Rectal Cancer: Staging, mrCRM, EMVI, Lymph Node Staging and Post-Treatment Response. Clin Colorectal Cancer. 2022;21(1):10–8.
Inoue A, Sheedy SP, Heiken JP, Mohammadinejad P, Graham RP, Lee HE, et al. MRI-detected extramural venous invasion of rectal cancer: Multimodality performance and implications at baseline imaging and after neoadjuvant therapy. Insights Imaging. 2021;12(1):110.
Shao K, Zheng R, Li A, Li X, Xu B. Clinical predictors of pathological good response in locally advanced rectal cancer. Radiat Oncol. 2021;16(1):10.
Tan Y, Fu D, Li D, Kong X, Jiang K, Chen L, et al. Predictors and Risk Factors of Pathologic Complete Response Following Neoadjuvant Chemoradiotherapy for Rectal Cancer: A Population-Based Analysis. Front Oncol. 2019;9:497.
Zhang Q, Liang J, Chen J, Mei S, Wang Z. Predictive Factors for Pathologic Complete Response Following Neoadjuvant Chemoradiotherapy for Rectal Cancer. Asian Pac J Cancer Prev. 2021;22(5):1607–11.
Feng L, Liu Z, Li C, Li Z, Lou X, Shao L, et al. Development and validation of a radiopathomics model to predict pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer: a multicentre observational study. Lancet Digit Health. 2022;4(1):e8–17.
Shahzadi I, Zwanenburg A, Lattermann A, Linge A, Baldus C, Peeken JC, et al. Analysis of MRI and CT-based radiomics features for personalized treatment in locally advanced rectal cancer and external validation of published radiomics models. Sci Rep. 2022;12(1):10192.
Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749–62.
Shin J, Seo N, Baek SE, Son NH, Lim JS, Kim NK, et al. MRI Radiomics Model Predicts Pathologic Complete Response of Rectal Cancer Following Chemoradiotherapy. Radiology. 2022;303(2):351–8.
Bordron A, Rio E, Badic B, Miranda O, Pradier O, Hatt M, et al. External Validation of a Radiomics Model for the Prediction of Complete Response to Neoadjuvant Chemoradiotherapy in Rectal Cancer. Cancers. 2022;14(4):1079.
van Timmeren JE, Cester D, Tanadini-Lang S, Alkadhi H, Baessler B. Radiomics in medical imaging—“how-to” guide and critical reflection. Insights Imaging. 2020;11(1):91.
Litvin AA, Burkin DA, Kropinov AA, Paramzin FN. Radiomics and Digital Image Texture Analysis in Oncology (Review). Mod Technol Med. 2021;13(2):97–104.
Alexandrescu ST, Dumitru AV, Babiuc RD, Costea RV. Assessment of clinical and pathological complete response after neoadjuvant chemoradiotherapy in rectal adenocarcinoma and its therapeutic implications. Rom J Morphol Embryol. 2021;62(2):411–25.
Gani C, Gani N, Zschaeck S, Eberle F, Schaeffeler N, Hehr T, et al. Organ Preservation in Rectal Cancer: The Patients’ Perspective. Front Oncol [Internet]. 2019 [cited 2023 Mar 22];9. Available from: https://www.frontiersin.org/articles/10.3389/fonc.2019.00318
Van Wickle JD, Paulson ES, Landry JC, Erickson BA, Hall WA. Adaptive radiation dose escalation in rectal adenocarcinoma: a review. J Gastrointest Oncol. 2017;8(5):902–14.
Yuval JB, Thompson HM, Garcia-Aguilar J. Organ Preservation in Rectal Cancer. J Gastrointest Surg Off J Soc Surg Aliment Tract. 2020;24(8):1880–8.
Chandramohan A, Siddiqi UM, Mittal R, Eapen A, Jesudason MR, Ram TS, et al. Diffusion weighted imaging improves diagnostic ability of MRI for determining complete response to neoadjuvant therapy in locally advanced rectal cancer. Eur J Radiol Open. 2020;7:100223.
Bibault JE, Giraud P, Housset M, Durdux C, Taieb J, Berger A, et al. Deep Learning and Radiomics predict complete response after neo-adjuvant chemoradiation for locally advanced rectal cancer. Sci Rep. 2018;8(1):12611.
Dekker A. Personal Health Train for Radiation Oncology in India and The Netherlands [Internet]. clinicaltrials.gov; 2020 Nov [cited 2021 Sep 6]. Report No.: NCT04655469. Available from: https://clinicaltrials.gov/ct2/show/NCT04655469
Mao Y, Pei Q, Fu Y, Liu H, Chen C, Li H, et al. Pre-Treatment Computed Tomography Radiomics for Predicting the Response to Neoadjuvant Chemoradiation in Locally Advanced Rectal Cancer: A Retrospective Study. Front Oncol. 2022;12:850774.
Gormly KL. High-Resolution T2-Weighted MRI to Evaluate Rectal Cancer: Why Variations Matter. Korean J Radiol. 2021;22(9):1475–80.
Xiang Y, Li S, Wang H, Song M, Hu K, Wang F, et al. MRI-based radiomics to predict neoadjuvant chemoradiotherapy outcomes in locally advanced rectal cancer: A multicenter study. Clin Transl Radiat Oncol. 2023;38:175–82.
Song M, Li S, Wang H, Hu K, Wang F, Teng H, 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.

The authors declare no competing interests.

SupplementryMaterials.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Can pre-treatment MRI and CT radiomics improve prediction of complete pathological response in locally advanced rectal cancer following neoadjuvant treatment?

Status:

Version 1

Abstract

Objective(s):

Methods

Results

Conclusion

Figures

Introduction

Materials and Methods

Patient cohort

Image acquisition

Tumour Delineation

Pathological Assessment

Radiomics Feature extraction

Experimental Set-up and Statistical Analysis:

Results

Patient characteristics

Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1