Radiomic Features Based on Multi-sequence MRI Predict Immunohistochemical Biomarkers of Endometrial Cancer

doi:10.21203/rs.3.rs-4179540/v1

Download PDF

Research Article

Radiomic Features Based on Multi-sequence MRI Predict Immunohistochemical Biomarkers of Endometrial Cancer

https://doi.org/10.21203/rs.3.rs-4179540/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Different molecular or genetic information influences the clinical decisions for patients diagnosed with endometrial cancer (EC). A non-invasive, precise, and efficient preoperative evaluation method is crucial for the prognosis of patients with EC.

Purpose: The aim of this study was to construct MRI-based radiomics models to predict immunohistochemical biomarkers and assess the relationship between radiomic features and the Ki-67 proliferation rate in EC.

Material and Methods: We retrospectively analyzed 100 estrogen receptor (ER), 94 progesterone receptor (PR), 97 P53, and 98 Ki-67 immunohistochemistry cases with EC who underwent magnetic resonance imaging (MRI) between May 2012 and June 2023 prior to surgery. Radiomic features were individually extracted from T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and the apparent diffusion coefficient (ADC). Least absolute shrinkage and selection operator (LASSO) regression was employed for feature selection. And logistic regression was employed to construct radiomics models with 5-fold cross-validation. The receiver operating characteristic (ROC) curves were analyzed to evaluate the performance of the radiomics models. Finally, Pearson's correlations were utilized to explore the association between the values of selected features and the Ki-67 proliferation rate.

Results: A total of 2264 features were extracted from each patient’s MRI sequences. The selected features from the multi-sequence models were shared with or without the single sequence models. Both single sequence and multi-sequence models demonstrated good diagnostic performance, although the diagnostic performance of multi-sequence models outperformed the single sequence models. Correlation analysis showed that adc_wavelet_glszm_wavelet-HHH-SmallAreaLowGrayLevelEmphasis and t2_log_firstorder_log-sigma-2-0-mm-3D-Skewness were negatively correlated with the Ki-67 proliferation rate.

Conclusions: MRI-based radiomic features are promising predictors of immunohistochemistry and prognosis in EC.

endometrial cancer

radiomics

immunohistochemical biomarkers

magnetic resonance imaging

correlation analysis

Endometrial cancer (EC) is a common gynecologic malignancy around the world. In recent years, the incidence rate of women in our country has increased [1, 2]. The significant heterogeneity of EC underscores the importance of accurate grading of EC for clinics [3, 4]. Although most early-stage EC patients have a good prognosis, a small number of patients have a poor prognosis even when diagnosed early [5]. It is because ECs with the same or similar histological characteristics may harbor different molecular or genetic information, which in turn influences the prognosis of patients [6].

In recent years, great progress has been made in understanding the molecular pathology of EC. A series of molecular markers have emerged, aiding in the clinical differential diagnosis and prognosis prediction of EC, among which estrogen receptor (ER), progesterone receptor (PR), P53, and Ki-67 are four vital biological behavior markers [7]. Notably, endometrioid adenocarcinoma (EEC) is a hormonally regulated disease, and a positive PR status is associated with a more favorable prognosis [8]. In addition, progestin therapy can be an option for selected patients, particularly young women who wish to preserve fertility. Conversely, double-negative hormone receptor (ER and PR) or negative PR status has been associated with shorter survival [9, 10]. The presence of P53 mutations is indicative of invasiveness and a poor prognosis, while the prognosis of patients with P53 wild type is better [11]. As for Ki-67, its expression was reported to be positively correlated with tumor grade in patients with EEC, and an increased risk of recurrence was found in Stage I-II EC patients with a higher Ki-67 index [12, 13]. Therefore, accurately identification of the molecular and genetic information in EC and precise treatment are important to clinical.

At present, the immunohistochemical types of EC are primarily obtained from the pathological histology after surgical resection or biopsy. However, it is an invasive procedure that highly dependent on the experience of the pathologists and sometimes prone to sampling biases [14, 15]. Magnetic resonance imaging (MRI) offers a non-invasive technique with excellent soft tissue contrast resolution. Conventional MRI, diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) maps are of great value in the diagnosis and staging of EC [16–18]. In recent years, the rapid advancements in artificial intelligence have given rise to the burgeoning field of radiomics. Many studies have used radiomics methods based on single or multi-sequence MRI to predict tumor grade, deep myometrial invasion (DMI), and lymph vascular space invasion (LVSI) in EC [19–21]. With the in-depth study of radiomics, it transcends the confines of tumor diagnosis and staging, delving deeper into the realm of histopathological features. Recently, Zijing Lin et al. developed a radiomics nomogram for the prediction of microsatellite instability (MSI) status in EC, demonstrating favorable calibration and clinical utility [22]. In addition, a single-center study used small data based on T2-weighted fluid-attenuated inversion recovery (T2 FLAIR) to predict immunohistochemical features of glioma, and satisfactory diagnosis accuracy was reported, indicating possible [23].

This study aimed to collect the immunohistochemical results of EC. By utilizing T2-weighted imaging (T2WI), DWI, and ADC images, the radiomics method was used to predict the immunohistochemical characteristics of EC through the extraction of radiomic features and construction of the radiomics model, which could contribute to a deeper understanding of the molecular characteristics of EC and provide new insights for guiding the precision treatment before clinical operation, improving the prognosis of patients.

Study participants

The institutional review boards of the Second Affiliated Hospital of Wenzhou Medical University approved this retrospective study, and individual consent for the retrospective analysis was waived. The study cohorts enrolled in this research followed the criteria as follows: (1) patients had clinical symptoms and was histologically diagnosed with EC; (2) availability of immunohistochemistry results; (3) no history of chemotherapy, radiotherapy, or surgery prior to the MRI examination; and (4) MRI examinations consisting of T2WI, DWI, and ADC maps of acceptable MR image quality. A total of 105 confirmed female patients (mean age, 55 years; age range, 27–74 years), who were pathologically diagnosed with EC and underwent MRI examinations at this center between May 2012 and June 2023, were reviewed in this study. Feature extraction or immunohistochemistry were not available for all; there were 100 ER, 94 PR, 97 P53, and 98 Ki-67 immunohistochemistry cases, respectively. The patients were randomly divided into a training set and a validation set according to a ratio of 8:2.

MRI acquisition

The MRI scans was performed using 3.0-T scanners with phased-array abdominal coils. The patients lay in a supine position and breathed freely during the acquisition. The following sequences were obtained: Sagittal T2WI with and without fat saturation; DWI with b = 800 or 1000 s/mm². The ADC map was automatically calculated based on DWI by the embedded software of the MRI equipment. More detailed information about these sequences is listed in Table 1.

Table 1

MRI examination’s parameters.
	Device 1	Device 2
MRI	3.0-T Discovery 750 GE	3.0-T HDXT GE
T2WI	TR/TE = 3475/69	TR/TE = 2900/73
	Matrix = 320×320	Matrix = 320×320
	thickness = 4mm	thickness = 4mm
	FOV = 280×380mm	FOV = 280×380mm
DWI	TR/TE = 2400/73	TR/TE = 5000/66
	Matrix = 192×160	Matrix = 192×160
	thickness = 5mm	thickness = 6mm
	FOV = 288×360mm	FOV = 380×380mm
	b = 1000	b = 1000
MRI, magnetic resonance imaging; T2WI, T2-weighted imaging; DWI, diffusion-weighted imaging; TE, echo time; TR, repetition time; FOV, field of view.

Immunohistochemical status assessment

All the patients involved in the current study underwent total hysterectomy, bilateral salpingo-oophorectomy, and/or pelvic/para-aortic lymphadenectomy. Immunohistochemical factors were evaluated in these patients through the acquisition of surgical samples. The specific procedures were carried out following the standard guidelines of the institution [24]. All immunohistochemical results were collected in the PACS system of the center. In the ER and PR evaluations, tumor cells were taken into account for densities, categorized as (-, ±, 1+, 2+, and 3+). In cases where the densities of tumor cells are ±, 1+,2+, and 3+, they were classified as positive (label = 1), whereas negative (label = 0). The expression level of the P53 and Ki-67 index was represented by the proportional staining (%). The proportional staining of P53 with 0% and 75–100% was considered mutant type (complete absence and overexpression, label = 1), and 1–75% was considered wild type (label = 0) [25]. And proportional staining of Ki-67 that equaled or was more than 50% was considered hyperproliferation (label = 0), and less than 50% was considered hypoproliferation (label = 1) [23].

Segmentation

The regions of interest (ROIs) that covered the whole lesion were manually segmented by two radiologists, each with 6 and 10 years of experience, using an open-source program called ITK-SNAP (version 3.8.0) on MRI images while blinded to the histopathological results. For each case, ROI was first drawn by one radiologist and then reviewed by the other radiologist to ensure high-quality final segmentation results. The manual drawing of each ROI was performed slice-by-slice on T2WI, DWI, and ADC and cross-referenced with each other, taking care to avoid including nearby normal myometrium or endometrium.

Feature extraction and selection

All ROIs from T2WI, DWI, and ADC maps, as well as the original MRI, were processed in batches by the uAI Data Assistant (United Imaging Intelligence, China). Subsequently, the radiomics module within the within the uAI Research Portal software (uRP) (United Imaging Intelligence, China) [26] was applied for feature extraction from each sequence, which complies with the standards set by the IBSI recommendation. And all extracted features were normalized using a Z-value normalization algorithm. Then, least absolute shrinkage and selection operator (LASSO) regression was used to reduce the dimension of features so as to obtain the optimized subset of features for constructing the final model. The LASSO analysis included choosing the regular parameter λ and determining the number of features. Once the number of features was determined, the most predictive feature subset was chosen, and the corresponding coefficients were evaluated. In this study, we aimed to compare the difference in predictive performance between the single-sequence model and the multi-sequence model, so we mixed the top 20 features from the three single-sequence models for each immunohistochemical parameter and conducted LASSO analysis again for feature dimensionality reduction to facilitate the subsequent establishment of the multi-sequence model. The process of radiomics analysis is shown in Fig. 1.

Model building and evaluation

Logistic regression was employed to construct radiomic models. In model training, 5-fold cross-validation was used for algorithm hyperparameter tuning based on the training set. Finally, receiver operating characteristic (ROC) curves were plotted. The area under the curve (AUC) and additional indices, including accuracy, sensitivity, and specificity, were calculated to evaluate the performance of the radiomics models.

Correlation analysis

Pearson's correlations were used to explore the correlation between the values of radiomic features selected from the multi-sequence model (Fig. 2D) and the Ki-67 proliferation rate. This work was performed on SPSS (version 26.0) and GraphPad Prism (version 9), with statistical significance set at p < 0.05. Due to condition, we did not calculate the correlation between the screened radiomic features and the expression of ER, PR, and P53.

Study participants

Among the 105 patients reviewed, 100 ER, 94 PR, 97 P53, and 98 Ki-67 immunohistochemistry cases were included, respectively. The EC was staged based on FIGO 2018. The histological subtype and histological characteristics of EC were determined from a surgical specimen or biopsy. Detailed patient demographics are presented in Table 2.

Table 2

The detailed demographic information of EC patients.
Parameters	ER	PR	P53	Ki-67
Age (mean ± SD, year)	55 ± 8	56 ± 7	55 ± 8	55 ± 8
Total number	100	94	97	98
Label
Label = 0	80 (80.00%)	67 (71.28%)	59 (60.82%)	35 (35.71%)
Label = 1	20 (20.00%)	27 (28.72%)	38 (39.18%)	63 (64.29%)
Histological subtype
EEC	85 (85.00%)	82 (87.23%)	83 (85.57%)	84 (85.71%)
Non-EEC	10 (1.00%)	9 (9.57%)	9 (9.28%)	9 (9.18%)
unknown	5 (5.00%)	3 (3.20%))	5 (5.15%)	5 (5.11%)
Histological grade
G1	29 (29.00%)	28 (29.79%)	28 (28.87%)	28 (28.57%)
G2	33 (33.00%)	32 (34.04%)	33 (34.02%)	33 (33.67%)
G3	16 (16.00%)	15 (15.96%)	16 (6.19%)	16 (16.33%)
unknown	22 (22.00%)	19 (20.21)	20 (20.62%)	21 (21.43%)
FIGO staging
I	89 (89.00%)	83 (88.30%)	86 (88.66%)	88 (89.80%)
II	2 (2.00%)	2 (2.13%)	2 (2.06%)	1 (1.02%)
III	7 (7.00%)	7 (7.44%)	7 (7.22%)	7 (7.14%)
IV	2 (2.00%)	2 (2.13%)	2 (2.06%)	2 (2.04%)
EC, Endometrial cancer; ER, estrogen receptor; PR, progesterone receptor; the positive, mutant type, and hypoproliferation groups were labeled 0; the negative, wide type, and hyperproliferation groups were labeled 1; EEC, endometrioid adenocarcinoma.

Feature extraction and selection

A total of 2264 features were extracted from each sequence (T2WI, DWI, and ADC maps) for each patient. Following feature selection, the most effective features that were retained by each model were illustrated in Fig. 2. The results showed that the selected features were different between single-sequence and multi-sequence, as follows:

In ER models, there were two features from the multi-sequence model shared with the ADC model (wavelet_firstorder_wavelet-LLL-Maximum and discretegaussian_glcm_ClusterShade); two features from the multi-sequence model shared with the DWI model (wavelet_glacm wavelet-LLH-ClusterShade and boxmean_firstorder_Skewness); and two features from the multi-sequence model shared with the T2WI model (wavelet_glszm_wavelet-HHL-SizeZoneNonUnifemityNormalized and wavelet_firstorder_wavelet-LHH-Skewness). Only one feature from the multi-sequence model shared without the ADC model (log_firstorder_log-sigma-1-0-mm-3D-Kurtosis).

In PR models, there were three features from the multi-sequence model shared with the ADC model (boxsigmaimage_glcm_DifferenceVariance, mean_glcm_CluterShade and wavelet_firstorder_wavelry-HLL-Skewness); one feature from the multi-sequence model shared with the DWI model (wavelet_ngtdm_wavelet-HHH-Contrast); and two features from the multi-sequence model shared with the T2WI model (log_firstorder_log-sigma-0-5-mm-3D-Kurtosis and wavelet_firstorder_wavelet-LHH-Skewness). And two features from the multi-sequence model shared without the ADC and DWI models (wavelet_glrlm_wavelet-HHL-ShortRunLowGraylevelEmphasis, selected from ADC maps; boxmean_glcm_ldm, selected from DWI).

In P53 models, there were three features from the multi-sequence model shared with the ADC model (wavelet_gldm_wavelet-HLH-LargeDependenceLowGrayLevelEmphasis, log_gldm_log-sigma-0-5-mm-3D-DependenceVariance and boxsigmaimage_glrlm_LowGrayLevelRunEmphasis); three features from the multi-sequence model shared with the DWI model (wavelet_glcm_wavelet-LLH-Contrast, wavelet_glszm_wavelet-HHL-SizeZnoeNonUniformityNomalilized and normalize_firstorder_Maximum); and two features from the multi-sequence model shared with the T2WI model (wavelet_firstorder_wavelet-HLL-Kurtosis and normalize_glrlm_ShortRunLowGrayLevelEmphasis). And no feature from multi-sequence model shared without the ADC, DWI, and T2WI models.

In Ki-67 models, there were three features from the multi-sequence model shared with the ADC model (boxmean_glcm_correlation, wavelet_glszm_wavelet-hhh-smallarealowgraylevelemphasis and gldm_log-sigma-4-0-mm-3d-largedependencehighgraylevelemphasis); no features from the multi-sequence model shared with the DWI model; and no features from the multi-sequence model shared with the T2WI model. And two features from the multi-sequence model not shared with ADC, DWI, and T2WI models (log_firstorder_log-sigma-2-0-mm-ed-maximum and, selected from DWIs; log_firstorder_log sigma-2-0-mm-3d-skewness and wavelet_dirstorder_wavelet-hhl-median, selected from T2WI). And no feature from the multi-sequence model shared without the ADC, DWI, and T2WI models.

Model performance

We proposed four multiple sequence models for revealing the immunohistochemical typing of ER, PR, P53, and Ki-67. The ROC curves of four models on both the training and validation sets are shown in Fig. 3. The performance of four multi-sequence models in predicting the immunohistochemical subtypes was shown in Fig. 4. And the accuracy, sensitivity, and specificity to quantify the predictive performance are listed in Table 3. The results showed that both single-sequence model and multi-sequence model performed well in predicting the four immunohistochemical parameters of EC with high accuracy. However, the AUC, accuracy, sensitivity, and specificity of the multi-sequence model were generally higher than those of the single-sequence model, which showed higher predictive performance.

Table 3

The predictive performance of sixteen models on different data sets.
	Sequence	Data Set	AUC (95% CI)	Sensitivity	Specificity	Accuracy
ER	T2WI	Training Set	0.838 (0.728–0.949)	0.625	0.881	0.830
		Validation Set	0.829 (0.568-1.000)	0.600	0.888	0.830
	DWI	Training Set	0.818 (0.718–0.919)	0.512	0.881	0.807
		Validation Set	0.804 (0.589–0.986)	0.500	0.850	0.780
	ADC	Training Set	0.887 (0.810–0.966)	0.812	0.809	0.810
		Validation Set	0.863 (0.678-1.000)	0.800	0.812	0.810
	Multi	Training Set	0.930 (0.874–0.990)	0.900	0.816	0.832
		Validation Set	0.894 (0.758–0.997)	0.850	0.812	0.820
PR	T2WI	Training Set	0.828 (0.730–0.926)	0.611	0.858	0.787
		Validation Set	0.820 (0.620–0.997)	0.660	0.849	0.799
	DWI	Training Set	0.863 (0.779–0.946)	0.647	0.840	0.785
		Validation Set	0.818 (0.619–0.992)	0.560	0.793	0.723
	ADC	Training Set	0.862 (0.759–0.967)	0.776	0.836	0.819
		Validation Set	0.821 (0.600-1.000)	0.707	0.795	0.767
	Multi	Training Set	0.918 (0.851–0.989)	0.778	0.884	0.854
		Validation Set	0.874 (0.733–0.993)	0.787	0.838	0.820
P53	T2WI	Training Set	0.825 (0.736–0.917)	0.717	0.725	0.722
		Validation Set	0.730 (0.490–0.952)	0.611	0.661	0.641
	DWI	Training Set	0.851 (0.769–0.937)	0.731	0.784	0.763
		Validation Set	0.771 (0.548–0.973)	0.739	0.744	0.742
	ADC	Training Set	0.848 (0.767–0.932)	0.757	0.754	0.755
		Validation Set	0.770 (0.569–0.954)	0.714	0.748	0.733
	Multi	Training Set	0.873 (0.800-0.951)	0.803	0.737	0.763
		Validation Set	0.833 (0.663–0.977)	0.768	0.747	0.755
Ki-67	T2WI	Training Set	0.709 (0.593–0.827)	0.687	0.579	0.648
		Validation Set	0.683 (0.429–0.929)	0.653	0.600	0.633
	DWI	Training Set	0.645 (0.515–0.776)	0.559	0.614	0.579
		Validation Set	0.602 (0.327–0.879)	0.554	0.571	0.560
	ADC	Training Set	0.795 (0.693-0.900)	0.627	0.750	0.671
		Validation Set	0.788 (0.572–0.976)	0.588	0.714	0.632
	Multi	Training Set	0.848 (0.756–0.943)	0.861	0.707	0.806
		Validation Set	0.802 (0.596–0.982)	0.840	0.600	0.755
AUC, area under the curve; CI, confidence interval; Multi, multiple sequences.

Correlation analysis assessments

The correlation analysis found that the value of adc_wavelet_glszm_wavelet-HHH-SmallAreaLowGrayLevelEmphasis (the prefix indicated that it was selected from ADC maps) was negatively correlated with Ki-67 proliferation rate (r = -0.221, p = 0.029, Fig. 5). And correlation analysis found that the value of t2_log_firstorder_log-sigma-2-0-mm-3D-Skewness (the prefix indicated that it was selected from T2WI) was negatively correlated with Ki-67 proliferation rate (r = -0.209, p = 0.040, Fig. 5).

In this study, we constructed twelve single-sequence and four multi-sequence models based on T2WI, DWI, and ADC maps for predicting immunohistochemical biomarkers in EC patients, with labels for ER, PR, P53, and Ki-67 as labels 0 or 1. Our results showed that the high order radiomic features are important potential predictors. Both single-sequence and multi-sequence models showed good predictive performance. Therefore, the radiomics models can provide a noninvasive and personalized management method for EC patients.

Our feature selection results showed that a significant proportion of features selected in the multi-sequence model were shared with those selected in the single-sequence model, although a few did not share features selected in the single-sequence model, which may be related to the weight of the radiomic features, and the repeatedly selected features may be important radiomic markers for predicting immunohistochemical features. Subsequent correlation analysis validated this hypothesis.

The results of our correlation study showed that there were two values of radiomic features that are negatively correlated with Ki-67 proliferation rate: adc_wavelet_glszm_wavelet-HHH-SmallAreaLowGrayLevelEmphasis and t2_log_firstorder_log-sigma-2-0-mm-3D-Skewness, respectively. Similar to previous studies exploring the relationship between the gene expression of caners and the values of radiomic features values, including breast cancer [27], lung cancer [28] and retinoblastoma[29], our findings showed that specific features were correlated with gene expression. For future research, we intend to further expand the sample size and even gene sequencing to explore the relationship between radiomic features and gene expression to further validate our hypothesis.

It is worth noting that our results showed that both single-sequence and multi-sequence models performed well in predicting immunohistochemical biomarkers with high accuracy. However, the prediction performance of the multi-sequence model was better than that of the single-sequence model. In previous radiomics studies on EC, we found that in two studies from the same team, the predictive power of the model for deep myometrial invasion (DMI), lympho-vascular space invasion (LVSI), and histologic high-grade EC improved after they added sequences [20, 30], which is consistent with our findings. This is because increasing the sequence can obtain more tumor information, and the radiomic features that rank highly in a single sequence are not necessarily effective compared to the radiomic features of other sequences. Adding sequences aims to capture the most effective radiomic features.

Our study had several limitations. Firstly, it difficult to obtain both MRI and postoperative immunohistochemical information for patients, resulting our current sample size was relatively small. Future cooperation with other institutions could help collect more retrospective and prospective samples to retain a more stable model. Secondly, dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) was not part of this study. In the scanning process, the image quality is different due to different MR equipment and technicians, and in clinical work, there are not many patients who undergo DCE-MRI scanning, so we plan to carry out prospective studies in the future to ensure the integrity of the data. Finally, we did not include clinical measures such as estrogen levels, BMI, and tumor indicators. We tried to collect this information to further improve the performance of our radiomic models, but in the medical record system, the indicators contained in the preoperative examination of patients were not exactly the same or some information was missing. Future inclusion of more samples is anticipated to enrich our experiment.

This study developed single-sequence and multi-sequence predicting models for four kinds of immunohistochemical biomarkers (ER, PR, P53, and Ki-67) in EC patients using T2WI, DWI, and ADC maps with good performance. Further correlation analysis showed an association between specific radiomic features and immunohistochemical biomarkers, providing a new way for clinicians to predict the molecular and gene information of patients preoperatively, contributing to advancing personalized precision medicine and improving the prognosis of patients.

endometrial cancer

estrogen receptor

progesterone receptor

EEC

endometrioid adenocarcinoma

MRI

Magnetic resonance imaging

DWI

diffusion-weighted imaging

ADC

apparent diffusion coefficient

DMI

deep myometrial invasion

LVSI

lymph vascular space invasion

T2 FLAIR

T2-weighted fluid-attenuated inversion recovery

T2WI

T2-weighted imaging

uRP

uAI Research Portal software

LASSO

least absolute shrinkage and selection operator

DCE-MRI

dynamic contrast-enhanced magnetic resonance imaging.

Ethics approval and consent to participate

The institutional review boards of the Second Affiliated Hospital of Wenzhou Medical University approved this retrospective study, and individual consent for the retrospective analysis was waived. We confirmed that the study was carried out in accordance with relevant guidelines and regulations of Helsinki Declaration.

Consent for publication

Not applicable.

Availability of data and materials

Not applicable.

Competing Interests

The authors declare that there is no conflict of interest.

Funding

This work was supported by grants from Wenzhou Science and Technology Bureau in China (No. Y2020816).

Authors' contributions

LS, XC: Drafting of the manuscript. LH, XW, ZY and JP: Critical revision of the manuscript. LS and YZ: Data analysis or interpretation. LS, LL and XC: Conception, Data acquisition. All authors: All authors reviewed and approved the manuscript.

Acknowledgements

Not applicable.

Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer Statistics, 2021. CA Cancer J Clin. 2021;71(1):7–33.
Crosbie EJ, Kitson SJ, McAlpine JN, Mukhopadhyay A, Powell ME, Singh N. Endometrial cancer. Lancet. 2022;399(10333):1412–28.
Jamieson A, Bosse T, McAlpine JN. The emerging role of molecular pathology in directing the systemic treatment of endometrial cancer. Ther Adv Med Oncol. 2021;13:17588359211035959.
Stelloo E, Nout RA, Osse EM, Jürgenliemk-Schulz IJ, Jobsen JJ, Lutgens LC, van der Steen-Banasik EM, Nijman HW, Putter H, Bosse T, et al. Improved Risk Assessment by Integrating Molecular and Clinicopathological Factors in Early-stage Endometrial Cancer-Combined Analysis of the PORTEC Cohorts. Clin Cancer Res. 2016;22(16):4215–24.
Tortorella L, Restaino S, Zannoni GF, Vizzielli G, Chiantera V, Cappuccio S, Gioè A, La Fera E, Dinoi G, Angelico G, et al. Substantial lymph-vascular space invasion (LVSI) as predictor of distant relapse and poor prognosis in low-risk early-stage endometrial cancer. J Gynecol Oncol. 2021;32(2):e11.
Pasanen A, Loukovaara M, Ahvenainen T, Vahteristo P, Bützow R. Differential impact of clinicopathological risk factors within the 2 largest ProMisE molecular subgroups of endometrial carcinoma. PLoS ONE. 2021;16(9):e0253472.
Jia M, Jiang P, Huang Z, Hu J, Deng Y, Hu Z. The combined ratio of estrogen, progesterone, Ki-67, and P53 to predict the recurrence of endometrial cancer. J Surg Oncol. 2020;122(8):1808–14.
Zhang Y, Zhao D, Gong C, Zhang F, He J, Zhang W, Zhao Y, Sun J. Prognostic role of hormone receptors in endometrial cancer: a systematic review and meta-analysis. World J Surg Oncol. 2015;13(1):208.
Huvila J, Talve L, Carpén O, Edqvist PH, Pontén F, Grénman S, Auranen A. Progesterone receptor negativity is an independent risk factor for relapse in patients with early stage endometrioid endometrial adenocarcinoma. Gynecol Oncol. 2013;130(3):463–9.
Guan J, Xie L, Luo X, Yang B, Zhang H, Zhu Q, Chen X. The prognostic significance of estrogen and progesterone receptors in grade I and II endometrioid endometrial adenocarcinoma: hormone receptors in risk stratification. J Gynecol Oncol. 2019;30(1):e13.
Vermij L, Léon-Castillo A, Singh N, Powell ME, Edmondson RJ, Genestie C, Khaw P, Pyman J, McLachlin CM, Ghatage P, et al. p53 immunohistochemistry in endometrial cancer: clinical and molecular correlates in the PORTEC-3 trial. Mod Pathol. 2022;35(10):1475–83.
Stefansson IM, Salvesen HB, Immervoll H, Akslen LA. Prognostic impact of histological grade and vascular invasion compared with tumour cell proliferation in endometrial carcinoma of endometrioid type. Histopathology. 2004;44(5):472–9.
Jiang P, Jia M, Hu J, Huang Z, Deng Y, Lai L, Ding S, Hu Z. Prognostic Value of Ki67 in Patients with Stage 1–2 Endometrial Cancer: Validation of the Cut-off Value of Ki67 as a Predictive Factor. Onco Targets Ther. 2020;13:10841–50.
Dueholm M, Hjorth IM. Structured imaging technique in the gynecologic office for the diagnosis of abnormal uterine bleeding. Best Pract Res Clin Obstet Gynaecol. 2017;40:23–43.
Di Spiezio Sardo A, De Angelis MC, Della Corte L, Carugno J, Zizolfi B, Guadagno E, Gencarelli A, Cecchi E, Simoncini T, Bifulco G, et al. Should endometrial biopsy under direct hysteroscopic visualization using the grasp technique become the new gold standard for the preoperative evaluation of the patient with endometrial cancer? Gynecol Oncol. 2020;158(2):347–53.
Lee Y, Kim KA, Song MJ, Park YS, Lee J, Choi JW, Lee CH. Multiparametric magnetic resonance imaging of endometrial polypoid lesions. Abdom Radiol (NY). 2020;45(11):3869–81.
Chen J, Fan W, Gu H, Wang Y, Liu Y, Chen X, Ren S, Wang Z. The value of the apparent diffusion coefficient in differentiating type II from type I endometrial carcinoma. Acta Radiol. 2021;62(7):959–65.
Bakir B, Sanli S, Bakir VL, Ayas S, Yildiz SO, Iyibozkurt AC, Kartal MG, Yavuz E. Role of diffusion weighted MRI in the differential diagnosis of endometrial cancer, polyp, hyperplasia, and physiological thickening. Clin Imaging. 2017;41:86–94.
Yan BC, Li Y, Ma FH, Zhang GF, Feng F, Sun MH, Lin GW, Qiang JW. Radiologists with MRI-based radiomics aids to predict the pelvic lymph node metastasis in endometrial cancer: a multicenter study. Eur Radiol. 2021;31(1):411–22.
Ueno Y, Forghani B, Forghani R, Dohan A, Zeng XZ, Chamming's F, Arseneau J, Fu L, Gilbert L, Gallix B, et al. Endometrial Carcinoma: MR Imaging-based Texture Model for Preoperative Risk Stratification-A Preliminary Analysis. Radiology. 2017;284(3):748–57.
Ytre-Hauge S, Dybvik JA, Lundervold A, Salvesen ØO, Krakstad C, Fasmer KE, Werner HM, Ganeshan B, Høivik E, Bjørge L, et al. Preoperative tumor texture analysis on MRI predicts high-risk disease and reduced survival in endometrial cancer. J Magn Reson Imaging. 2018;48(6):1637–47.
Lin Z, Wang T, Li H, Xiao M, Ma X, Gu Y, Qiang J. Magnetic resonance-based radiomics nomogram for predicting microsatellite instability status in endometrial cancer. Quant Imaging Med Surg. 2023;13(1):108–20.
Li J, Liu S, Qin Y, Zhang Y, Wang N, Liu H. High-order radiomics features based on T2 FLAIR MRI predict multiple glioma immunohistochemical features: A more precise and personalized gliomas management. PLoS ONE. 2020;15(1):e0227703.
Yu X, Guo S, Song W, Xiang T, Yang C, Tao K, Zhou L, Cao Y, Liu S. Estrogen receptor α (ERα) status evaluation using RNAscope in situ hybridization: a reliable and complementary method for IHC in breast cancer tissues. Hum Pathol. 2017;61:121–9.
Singh N, Piskorz AM, Bosse T, Jimenez-Linan M, Rous B, Brenton JD, Gilks CB, Köbel M. p53 immunohistochemistry is an accurate surrogate for TP53 mutational analysis in endometrial carcinoma biopsies. J Pathol. 2020;250(3):336–45.
Wu J, Xia Y, Wang X, Wei Y, Liu A, Innanje A, Zheng M, Chen L, Shi J, Wang L, et al. uRP: An integrated research platform for one-stop analysis of medical images. Front Radiol. 2023;3:1153784.
Kim GR, Ku YJ, Kim JH, Kim EK. Correlation between MR Image-Based Radiomics Features and Risk Scores Associated with Gene Expression Profiles in Breast Cancer. Taehan Yongsang Uihakhoe Chi. 2020;81(3):632–43.
Wang T, Gong J, Duan HH, Wang LJ, Ye XD, Nie SD. Correlation between CT based radiomics features and gene expression data in non-small cell lung cancer. J Xray Sci Technol. 2019;27(5):773–803.
Jansen RW, Roohollahi K, Uner OE, de Jong Y, de Bloeme CM, Göricke S, Sirin S, Maeder P, Galluzzi P, Brisse HJ et al. Correlation of gene expression with magnetic resonance imaging features of retinoblastoma: a multi-center radiogenomics validation study. Eur Radiol 2023.
Lefebvre TL, Ueno Y, Dohan A, Chatterjee A, Vallières M, Winter-Reinhold E, Saif S, Levesque IR, Zeng XZ, Forghani R, et al. Development and Validation of Multiparametric MRI-based Radiomics Models for Preoperative Risk Stratification of Endometrial Cancer. Radiology. 2022;305(2):375–86.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Radiomic Features Based on Multi-sequence MRI Predict Immunohistochemical Biomarkers of Endometrial Cancer

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Study participants

MRI acquisition

Immunohistochemical status assessment

Segmentation

Feature extraction and selection

Model building and evaluation

Correlation analysis

Result

Study participants

Feature extraction and selection

Model performance

Correlation analysis assessments

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1