Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: In recent years,many neural network models based on prostate lesion segmentation in magnetic resonance images have different stability and diagnostic efficiency.

Goal(s): we want to get an automatic segmentation model with high performance for the prostate and its lesion region.

Approach: Our Network DCNN is inspired by the U-Net model with the encoding-decoding path as the backbone,importing dense block,attention mechanism techniques,and group norm-Atrous Spatial Pyramidal Pooling,these could be broadly used to improve the capability of CNN.

Results: Compared to the state-of-the-art models,FCN,U-Net,U-Net++,and ResUNet.The segmentation performance of DCNN for prostate lesions On the MR DWI image swas better than the other models.

Impact: The DCNN model with dense block, convolution block attention module, and group norm-Atrous Spatial Pyramid Pooling performed well in the segmentation of the prostate and its lesion regions. which supports its potential to assist prostate disease diagnosis in clinical medicine.

Keywords: Diagnosis/Prediction, Radiomics, hepatic fibrosis，Gd-EOB-DTPA，hepatobiliary phase

Motivation: Noninvasive assessment of hepatic fibrosis progression in individual patients.

Goal(s): To explore the value of signal intensity ratio (SIR) and radiomics analysis based on hepatobiliary phase (HBP) of Gd-EOB-DTPA-enhanced MRI in evaluating hepatic fibrosis staging.

Approach: The liver–muscle contrast ratio (LMCR) and liver–spleen contrast ratio (LSCR) were measured in HBP image, and the radiomics model and SIR–radiomics combined model were established. Finally, the diagnostic accuracy of SIR and both models in hepatic fibrosis staging were evaluated.

Results: The radiomics model exhibited higher diagnostic efficacy than SIR. The SIR–radiomics combined model demonstrated superior diagnostic performance in evaluating hepatic fibrosis staging .

Impact: This study constructed a noninvasive hepatic fibrosis diagnostic model using accessible routine sequence and improved the diagnostic efficiency. It offers a reliable approach for early diagnosis and treatment assessment, potentially reducing the need for unnecessary invasive biopsies in the future.

Keywords: Diagnosis/Prediction, Tumor, Endometrial cancer; habitats; radiomics; prediction model; molecular subtype

Motivation: Endometrial cancer (EC) is a highly heterogeneous cancer comprising of both histological and molecular subtypes. The p53abn subtype is associated with a poor prognosis particularly. 

Goal(s): We aimed to develop habitats based multiparametric MRI radiomics model for the prediction of EC molecular subtype and evaluated the performance.

Approach: Our study is a dual-center retrospective research.

Results: Our habitats model demonstrated good performance in both internal and external validations. It exhibited higher efficacy compared to radiomics and clinical models.

Impact: Using a non-invasive modality method to trigger these subtypes of EC as early as possible will aid clinicians to establish individual treatment. This research also marks the first use of habitat analysis in the study of EC.

Keywords: Diagnosis/Prediction, Radiomics

Motivation: There is no appropriate index to assess the prognosis of patients with endometrial cancer after surgery.

Goal(s): Our goal was to explore the value of an MRI-based radiomics nomogram in predicting the progression-free survival (PFS) of patients with endometrial cancer.

Approach: A nomogram was established using multivariable Cox regression by incorporating significant clinical-pathological predictors and radiomics signatures developed by another least absolute shrinkage and selection operator Cox regression.

Results: The nomogram proved itself valuable in predicting PFS. Furthermore, the radiomics nomogram–defined risk stratification was associated with PFS.

Impact: The novel nomogram may contribute to the precise stratification of patients with a high risk of progression. Therefore, more active follow-up and adjuvant therapy should be carried out after surgery to prolong and improve the patients' quality of life.

Keywords: Diagnosis/Prediction, Tumor, Endometrial cancer; Lymphatic metastasis; Lymph node; Magnetic resonance imaging; Radiomics

Motivation: Whether lymph node metastasis (LNM) affects surgical management in endometrial cancer (EC) patients.

Goal(s): To develop and validate a nomogram based on intra- and peritumoral radiomics features and multiparameter MRI imaging features to preoperatively predict LNM in EC.

Approach: Three hundred and seventy-four women with histologically confirmed EC were divided into training (n = 220), test (n = 94), and external-validation cohorts (n = 60). Radiomics features were extracted from intra- and peritumoral regions based on axial T2WI and ADC mapping. 

Results: The nomogram combining the Radscore, CA125, and TAR showed good diagnostic performance (AUC_training = 0.878, AUC_test = 0.877, AUC_{external-validation} = 0.864).

Impact: The combined intra- and peritumoral region multiparameter MRI radiomics nomogram could be used to preoperatively predict LNM in EC. Moreover, different field strength data were proportionally mixed for modeling and external validation, expanding the real-world application scenarios of prediction models.

Keywords: Diagnosis/Prediction, Radiomics, MR-HIFU, Treatment of Uterine Fibroids, XGBoost, SHAP, LR, Coefficients

Motivation: MR-HIFU offers a new treatment option for women with uterine fibroids. However, there is currently a lack of quantitative models to predict the efficacy of MR-HIFU based on T2WI of fibroids for guiding preoperative clinical decisions.

Goal(s): We hope to identify the most important predictive factors of MR-HIFU treatment for uterine fibroids and predict the efficacy using radiomics data combine with clinical data.

Approach: We employed XGBoost and logistic regression (LR) to build two prediction models.  SHAP values of XGBoost and LR coefficients were used to pinpoint significant predictive factors.

Results: Both models achieved outstanding results and the significant predictive factors are consistent.

Impact: Our excellent model results have identified the optimal predictive factors for assessing the efficacy of MR-HIFU in the treatment of uterine fibroids. These factors aid physicians in preoperative guidance and clinical strategy formulation, clarifying which patients will achieve better outcomes.

Keywords: Diagnosis/Prediction, Cancer

Motivation: Preoperational identification of lymph node metastasis (LNM) and lymphatic vascular space invasion (LVSI) of endometrial cancer from MRI is important to treatment planning.

Goal(s): To explore power of intra/peri-tumor radiomic features from DWI, T1CE and T2W images to identify LVSI and LNM.

Approach: We developed radiomics models with intra/peri-tumor features from different MRI images and compared their performance.We developed radiomics models for intra- and peri-tumoral features and compare performance.

Results: For LVSI, T2W model using both intra- and peri-tumoral features achieved AUC values of 0.790/0.696 in internal/external test cohorts. For LNM, the combined model achieved AUC values of 0.801/0.976 in internal/external test cohorts.

Impact: The radiomics signatures built with intra- and peri-tumoral features extracted from DWI, T1CE, T2W sequences can yield satisfactory predictions for both LVSI and LNM status in endometrial cancer.

Keywords: Diagnosis/Prediction, Prostate, magnetic resonance imaging, radiomics, tumor-stroma ratio, tumor microenvironment

Motivation: Tumor stroma is considered one of the key participants in prostate cancer development, progression, and even treatment resistance as an independent predictor, is associated with aggressiveness in a variety of malignancies.

Goal(s): We would like to apply the value of stroma cells in clinical practice for assessing the aggressiveness of PCa.

Approach: Five multiparametric magnetic resonance imaging (mp- MRI) radiomics feature-based machine learning models were developed and assessed to predict the tumor-stroma ratio (TSR) of PCa.

Results: The developed Multi-Layer Perception model showed excellent performance at predictive the TSR in prostate cancer with the area under the ROC curve (AUC) at 0.860.

Impact: This study constructed a mp-MRI-based radiomics model which is capable of accurately predicting the TSR of PCa and may serve as a complementary tool for assisting in risk stratification and guiding treatment decisions.

Keywords: Diagnosis/Prediction, DSC & DCE Perfusion, Prostate, DCE

Motivation: Existing techniques for DCE-MRI analysis is time-consuming and often assume a fixed arterial input function (AIF) across various locations and patients, leading to imprecise outcomes.

Goal(s): 1) The development of a Deep learning model for fast DCE-MRI analysis, and 2) the design of location- and patient-specific AIFs.

Approach: We use deep-learning model for fast DCE-MRI analysis, and propose to represent dispersion-applied AIF to allow for location- and subject-specific AIFs by interpolation between constant AIFs.

Results: 1) Reduced per-patient processing time by one-tenth, 2) improved fitting accuracy, and 3) higher-contrast parameteric maps between the lesion and normal tisue.

Impact: Other scientists, clinicians and patients may benefit from the faster processing time, and higher-contrast parametric maps for cancer diagnosis.

Keywords: Diagnosis/Prediction, Uterus, endometrial cancer; lymph node; nomogram

Motivation: Routinely MRI can’t assess lymph node metastasis (LNM) status in endometrial cancer (EC) accurately. 

Goal(s):  This study aimed to investigate whether the MRI radiomics model can predict LNM status in EC. 

Approach: 313 EC patients who underwent preoperative MRI were recruited with the status of LNM confirmed by pathology. 2880 radiomic features were extracted and three models including a clinical model, a radiomics model and a combined model were built.

Results:  Area under the receiver operating characteristic curves showed all of the three models can predict LNM and the combined model showed the best discrimination ability on the training set and test set.

Impact: This study built an accurate preoperative and non-invasive model to assess lymph node metastasis status in endometrial cancer. It could help doctors to determine the extent of lymphadenectomy and provide useful information for prognosis prediction and treatment decision.

Keywords: Diagnosis/Prediction, Radiomics, Deep learning

Motivation: Cervical stromal invasion (CSI) plays a critical role in distinguishing between stage I and II endometrial carcinoma (EC) and serves as a key prognostic indicator.

Goal(s): Assisting clinicians in achieving precise preoperative treatment and prognostic assessments.

Approach: This study constructed innovative machine learning models that merge radiomics and 3D deep transfer learning to preoperatively and non-invasively predict CSI.

Results: Novel machine learning model has significant superiority over radiologists for preoperative prediction of CSI.

Impact: Constructing a non-invasive preoperative prediction model to increase the diagnostic accuracy of CSI, makes up for the limitations of traditional imaging observation in the assessment of CSI and subsequently directs clinicians in preoperative precise treatment and prognostic evaluation.

Keywords: Diagnosis/Prediction, Prostate

Motivation: Biochemical recurrence (BCR) remains a significant concern for patients after radical prostatectomy.

Goal(s): This study’s objective was to determine whether automated mpMRI cancer risk maps could predict postsurgical prostate cancer (PCa) progression (Biochemical Recurrence (BCR).

Approach: Derived lesion volumes and mpMRI parameters from the cancer risk maps were used to analyze factors for BCR.

Results: A decision tree model for BCR was generated with sensitivity = 0.81, specificity = 0.82, and ROC AUC = 0.85. The prediction results indicate the potential of mpMRI and PCa risk maps to improve prediction of BCR after prostatectomy.

Impact: In this study we used preprostatectomy multiparametric MRI (mpMRI) cancer risk maps to analyze potential predictors of postsurgical prostate cancer (PCa) progression (Biochemical Recurrence (BCR) or treatment failure). A decision tree model was generated with ROC AUC 0.85.

Keywords: Diagnosis/Prediction, Cancer, deep learning, diffusion weighted imaging, cervix cancer, grade, stage

Motivation: Conventional grading of cervix cancer (CC) requires biopsy that can lead to potential side-effects. Radiologists typically rely on multi-imaging for CC staging, which can result in patient discomfort, higher costs, and increased workload.

Goal(s): This study aims to introduce noninvasive methods that leverage a single MRI for both grading and staging prediction.

Approach: EfficientNetB0 and EfficientNetB3 were applied for tumor classification (binary and four-class) based on apparent diffusion coefficient maps of 85 patients. They were evaluated using the area under the receiver operating characteristic curve (AUC).

Results: High AUC=0.924 and AUC=0.931 were obtained for grade and stage predictions respectively.

Impact: The results demonstrated the feasibility of noninvasive prediction of cervical cancer grade and stage from diffusion weighted images. This could significantly impact the diagnosis and management of cervical cancer, as it can provide valuable information without biopsy or extensive imaging.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: The treatment methods and prognosis of cellular uterine leiomyomas (CULs) and uterine sarcomas (USs) are different. The ADC values has certain differential diagnostic value, but there is some overlap between them. Texture analysis (TA) may have some potential and complementary role in differential diagnosis.

Goal(s): To explore the capability of TA based on MRI and ADC values in the differential diagnosis of USs from CULs.

Approach: Combining the ADC values and texture parameters to set up diagnostic model and evaluate the diagnosis value and clinical usefulness of the model.

Results: Texture analysis combined with DWI could be helpful to distinguish USs and CULs.

Impact: Texture analysis combined with DWI give a better method to identify uterine sarcomas and cellular uterine leiomyomas, providing a more reliable basis for the choice of clinical treatment.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Rectal Cancer

Motivation: To explore and validate the association between magnetic resonance image texture features and the efficacy of Neoadjuvant Chemoradiotherapy for rectal cancer.

Goal(s): To predict the efficacy of neoadjuvant chemoradiotherapy for locally advanced rectal cancer using machine learning methods.

Approach: The wavelet texture parameters of all lesions in patients' MRI images were extracted, and feature selection was performed using random forest classifier model, and then classification learning was performed using the XGBoost classifier.

Results: The model based on wavelet texture feature analysis of MRI can effectively predict the effect of neoadjuvant radiochemotherapy for rectal cancer patients.

Impact: Through this study, we explored and validated the relationship between MRI texture features and the efficacy of Neoadjuvant Chemoradiotherapy for rectal cancer, providing new guidance and decision support for individualized treatment strategies.

Keywords: Diagnosis/Prediction, Segmentation

Motivation:  A significant challenge in RC management is the precise delineation of tumor boundaries and the accurate evaluation of LNs. The existing manual processes for this are time-consuming and subject to high variability.

Goal(s):   This study aimed to develop a deep learning approach for simultaneous RC tumor and LN segmentation.

Approach: We constructed the model with mpMRI data input, ResUNet architecture, and focal cross entropy loss.

Results: The ResUNet model achieved a mean SEN of 0.824, PRE of 0.619, and DSC of 0.694 in the validation dataset, indicating promising results. However, some false positives and false negatives were observed in LN segmentation.

Impact: We introduced a ResUNet model for RC segmentation and achieved satisfactory results. While our findings are preliminary and may benefit from larger samples, this approach could improve tumor and LN segmentation and ultimately enhance clinical utility in RC management.