Keywords: Diagnosis/Prediction, Radiomics, Rectal cancer; Neoadjuvant chemoradiotherapy; Distant metastasis; Pathomics; Immunosocre

Motivation: Identifying high-risk patients for distant metastasis (DM) before treatment can facilitate the development of personalized neoadjuvant treatment and improve the prognosis of patients with locally advanced rectal cancer (LARC). 

Goal(s): This study aimed to construct a predictive model that integrates radiological information at the macroscale and pathological information at the microscale to estimate the probability of DM in LARC patients after neoadjuvant chemoradiotherapy, using radiomics, pathomics, and biopsy-adapted immunoscore. 

Approach: Feature selection and signature construction were performed using the least absolute shrinkage and selection operator (LASSO)-Cox analysis.

Results: The results demonstrated the effectiveness of the nomogram in identifying high-risk DM patients.

Impact: Incorporating multiscale information, including radiomics, pathomics, and the immune microenvironment, enhances the characterization of tumors and provides a robust model for identifying high-risk DM patients in LARC. This approach aids in the development of personalized neoadjuvant treatment strategies.

Keywords: Diagnosis/Prediction, Cancer

Motivation: Neoadjuvant systemic therapy (NAST) followed by surgery is the standard of care for triple-negative breast cancer (TNBC) patients. However, only approximately half of these patients achieve pathological complete response (pCR). 

Goal(s): To build a prediction model to identify non-pCR patients before the initiation of NAST.

Approach: We evaluated multiple prediction models using pretreatment multi-parametric MRI from a cohort of 282 TNBC patients. 

Results: Our findings revealed that combined with clinical information, the best-performing model achieved an AUC of 0.74 on an independent testing set. We further observed that the performance of our models is not sensitive to the voxel selections in tumor segmentation.

Impact: Deep learning models for predicting pathological complete response to neoadjuvant systemic therapy of triple-negative breast cancer were developed using baseline multi-parametric MRI data and clinical information and achieved an AUC of 0.74 on the independent testing dataset.

Keywords: Diagnosis/Prediction, Pelvis, Imaging phenotype

Motivation: Heterogeneity of endometrial carcinoma (EC) leads to differences in prognosis among different patients. The method of unsupervised machine learning can classify tumors into different subtypes by identifying heterogeneity and similarity in radiomics features, which may have the ability of preoperative risk stratification.

Goal(s): To identify the intrinsic imaging phenotype for EC using multi-modality MR-based radiomics features.

Approach: Ten multi-omics clustering methods were used for imaging phenotypes identification and reached a consensus.

Results: Among the three identified imaging phenotypes, multiple pathological features and disease-free survival time showed significant differences.

Impact: Based on multi-modality MRI using an unsupervised machine learning approach to classify EC into different imaging phenotypes, which were associated with clinicopathological features and prognosis, and can be used for preoperative risk stratification.

Keywords: Diagnosis/Prediction, Radiomics, Brain metastasis

Motivation: Control of metastatic and primary tumors has been identified as prognostic factors for lung cancer patients with brain metastasis. However, prognosis prediction by combining imaging features of metastatic and primary tumors was less explored.

Goal(s): This study investigated the prediction efficacy based on image traits of brain metastasis and primary lung cancer.

Approach: The radiomic features separately extracted from brain MRI and chest CT images were merged to build the survival prediction models. 

Results: The proposed prediction model showed superior performance compared to the models based on a single modality in lung cancer with brain metastasis.

Impact: This study suggested that survival prediction can be enhanced by combining features of brain metastasis MRI and lung cancer CT. Imaging characteristics of both primary and secondary (metastatic) tumors are valuable for prognostic prediction in lung cancer with brain metastasis.

Keywords: Diagnosis/Prediction, Cancer

Motivation: Accurate selection of prostate cancer patients to undergo active surveillance (AS) is crucial to ensure suitable treatment.

Goal(s): To develop an automated framework for mpMRI analysis to assist clinical decision making about whether a patient should remain on AS.

Approach: We developed a progression risk stratification model using mpMRI data from an AS trial, and incorporating clinical biomarkers and radiomic features from lesions identified by a deep neural network.

Results: The lesion segmentation network achieved a median DSC of 60.7%, and the progression prediction model an AUC of 81.1% in determining likelihood of progression within 12 months.

Impact: We present a fully automated methodology to assess prostate cancer progression risk for AS patients within the timeframe between their follow-up visits, thereby providing essential data for clinicians that can prospectively improve AS patient selection.

Keywords: Diagnosis/Prediction, Cancer

Motivation: Accurate prediction of pathological complete response (pCR) after neoadjuvant chemotherapy enables individualized treatment options to avoid unnecessary breast excision and improve patients’ life quality.

Goal(s): To improve the prediction accuracy by simultaneously extracting temporal and spatial features of MRI signal during contrast enhancement.

Approach: A histogram signature is designed by concatenating histograms at different enhancing phases into a 2D picture and classified by convolutional neural network into pCR or non-pCR.

Results: The AUC, sensitivity, specificity of the histogram signature for pCR prediction is 0.833 in the test group (n=132). The model combining histogram signature with ER and HER2 increases AUC to 0.842. 

Impact: Histogram signatures from multi-phase MRI can be used as a new marker to measure tumor heterogeneity, estimate drug uptake, evaluate treatment response and predict prognosis for breast cancer or other cancers.

Keywords: Diagnosis/Prediction, Radiomics

Motivation: A precise assessment of LN restaging following nCRT is important to guide therapeutic decision and predict prognosis for LARC patients.

Goal(s): To develop and validate a predictive radiomics model for assessing LNM status after nCRT in LARC.

Approach: This study enrolled 150 LARC patients from two centers and constructed several radiomics models based on T2WI and DWI before or/and after nCRT to assess LNM after nCRT.

Results: The multiparametric model incorporating MR radiomics features prior to and after nCRT was superior to the clinical model, model_{pre_T2_DWI} and the single-sequence models (external validation cohort AUC 0.831).

Impact: Radiomics analysis of pre- and post-nCRT multiparameter MR images could predict LNM after nCRT in patients with LARC, and might help guide therapies and predict prognosis for LARC patients. 

Keywords: Diagnosis/Prediction, Radiomics

Motivation: Predicting LVSI before surgery remains a critical challenge.

Goal(s): To predict preoperative LVSI in patients with endometrial cancer in a noninvasive way.

Approach: We developed and validated MRI radiomics and clinical-radiomics models based on the features extracted from tumor and peritumoral regions.

Results: The clinical-radiomics model based on the features extracted from the tumors with 3mm peritumoral region exhibited the hightest predictive performance in the training cohort and testing cohort with an AUC of 0.86 and 0.86, respectively. The model also displayed clinical validity as depicted in the DCA curve.

Impact: Incorporating the radiomics features extracted from tumor with 3mm peritumoral region and clinical significance factors can improve the predictive efficacy of the model for predicting LVSI and increase its applicability in clinical practice.

Keywords: Diagnosis/Prediction, Radiomics, Explainability, Analysis/Processing, Cancer, Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Prostate, Software Tools

Motivation: Clinical use of computer-aided diagnosis systems for prostate cancer is currently hindered by their internal complexity. Explainability tools can give insight into the functioning of these machine learning (ML) models.

Goal(s): Our goal was to supplement the predictions of an MRI radiomics-based ML model for prostate cancer detection with explanations based on clinical concepts currently used in radiological assessment.

Approach: We clustered correlating MRI radiomics features into groups representing clinical concepts underlying the PI-RADS system. We used SHAP analysis to explain the importance of these concepts in each predicted lesion.

Results: Explainability based on clinical concepts gives insight into ML model predictions.

Impact: Our machine learning pipeline combines accurate prostate cancer detection on MRI with intrinsic explainability, potentially resulting in an easier integration into clinical use. 

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, lung cancer patients with brain metastasis

Motivation: To determine an effective individualized treatment decision for lung cancer patients with brain metastasis (BrM) to receive targeted therapy.

Goal(s):  we developed an MRI-based nomogram that can predict the response of lung cancer BrM to targeted therapy using multi-center data. 

Approach: Clinical predictors, radiomics and deep learning features extracted from BrM baseline MR images were incorporated to establish the nomogram using the LASSO logistics coefficients.

Results: The nomogram can accurately predict the 6-month and 12-month responses of BrM to targeted therapy across the training cohort, internal validation cohort, and external test set, outperforming all other models.

Impact: The MRI-based nomogram can be used as a pretreatment and personalized tool to predict response to targeted therapy in lung cancer patients with BrMs and thus assist in optimizing treatment for lung cancer patients who suffer from BrMs.

Keywords: Diagnosis/Prediction, Breast

Motivation: Her2-low breast cancers could benefit from new anti-HER2 therapies.  

Goal(s): To construct a preoperative prediction model of HER2 expression levels using multiparametric MRI and machine learning (ML) algorithms.

Approach: 621 patients were investigated. Four ML methods were used to build models based on MRI features to predict HER2 expression levels.

Results: MRI features of multiple lesions, spiculated margin, peritumoral edema and largest diameter were selected to build the models. ML models performed better for predicting HER2-zero vs. HER2-low/-overexpression than HER2-low vs. HER2-overexpression. The best model was KNN of AUC 0.86, sensitivity of 76%, specificity of 73%, and accuracy of 75%.

Impact: MRI features of breast cancer are associated with different HER2 expression levels. MRI-based ML models have the potential to preoperatively predict the HER2 expression status.

Keywords: Diagnosis/Prediction, Pelvis

Motivation: High-grade serous ovarian carcinoma (HGSOC) poses a significant challenge due to platinum resistance and the inherent difficulty in its prediction.

Goal(s): We aimed to explore MRI-based habitat model for predicting response of platinum-based chemotherapy in HGSOC patients, and compared with radiomics and deep learning models.

Approach: We leveraged the K-means algorithm for clustering on multiparameter MRI data. Then the radiomics, habitat, and deep learning models were constructed.

Results: Habitat model had the potential to predict platinum resistence, with a superior performance to radiomics and deep learning models. The nomogram integrating habitat with neoadjuvant chemotherapy yielded a better performance compared to others.

Impact: This study holds substantial clinical significance as it establishes a foundational framework for the customization of treatment strategies for patients afflicted with HGSOC.

Keywords: Diagnosis/Prediction, Cancer

Motivation: Millions of prostate biopsies are being ordered each year, a great majority of which yield negative results. A reliable and non-invasive method for detecting prostate cancer grade is critical.

Goal(s): To develop a robust and efficient MRI-based non-invasive model to detect the Gleason score of the lesions without the need for a biopsy. 

Approach: We propose a physics-informed autoencoder that integrates the strengths of model-based and deep learning-based methods, while overcoming their respective weaknesses.

Results: Physically-interpretable biomarkers that our model yields correlate strongly with Gleason score, providing important new diagnostic markers, and laying the groundwork for a potential new quantitative MRI method.

Impact: The proposed model offers for many potential usages in diagnostic radiology, by presenting a non-invasive method for diagnosing and staging prostate cancer, potentially affecting about a million patients annually by reducing unnecessary biopsies and saving millions in healthcare costs. 

Keywords: Diagnosis/Prediction, PET/MR, Glioblastoma, WSI

Motivation: Glioblastoma (GBM) is a fast-growing invasive brain tumor that presents unique treatment challenges. Early diagnosis requires manual segmentation using MRI and histopathological image analysis.

Goal(s): Our proposed model can facilitate medical personnel in an efficient and accurate diagnosis of glioblastoma.

Approach: We present a multiscale multilevel approach based on deep learning for precision medicine and quantitative imaging in GBM capturing image feature and providing wide-ranging contextual information.

Results: Our method predicted the overall survival of GMB patients with an average accuracy of 88.63% and 91.7% DSC (Unet: 84% DSC; Swin Transformer: 87% DSC) on BraTS 2020.

Impact: Our model surpasses state-of-the-art methods in Glioblastoma (GBM) segmentation and predicts patient survival with 88.63% accuracy. This research work assists in precise and efficient diagnoses of GBM, potentially contributing to early disease detection and treatment strategies.

Keywords: Diagnosis/Prediction, Data Processing

Motivation: Molecular markers, such as IDH and TERTp, have been reported as significant prognostic factors in gliomas.

Goal(s): The aim of this study is to predict IDH and TERTp mutational subtypes in gliomas non-invasively using deep-learning applied to rCBV images derived from DSC-MRI.

Approach: We proposed a deep-learning approach with attention gates to classify IDH- and TERTp-mutation subgroups of gliomas using rCBV images along with anatomical-MRI. Additionally, Grad-CAM approach was employed to provide an explanation of which image sections played a role in decision-making.

Results: Attention-boosted deep learning-based classification model yielded high accuracy rates. GradCAM approach also highlighted the significance of different tumor components.

Impact: The proposed attention-boosted deep learning based method might have the potential to assist clinicians in the noninvasive identification of IDH and TERTp mutations at the pre-surgery point and potentially enhance treatment strategies and patient outcomes.

Keywords: Diagnosis/Prediction, Brain

Motivation: Deep learning networks offers an opportunity for diffuse gliomas classification, which may be help for therapeutic decision making and selection of patient groups suitable for targeted genetic analysis.

Goal(s): The purpose of this study is to develop an artificial intelligence method to reclassify adult-type diffuse gliomas based on the new WHO CNS tumor classification. 

Approach: An artificial intelligence decision tree diagnostic platform(DTDP) based on MRI and deep learning networks was developed by combined 6 individualized CNNs models in series and parallel 

Results:  The DTDP performed well with accuracy of 86.67%.

Impact: The DTDP achieved automatic classification and comprehensive diagnosis of adult‑type diffuse gliomas by combining genetic biomarkers and histological grading, and effectively helped neuroradiologists to reclassify adult-type diffuse gliomas.

Keywords: Diagnosis/Prediction, Radiomics, Gliomas

Motivation: The MGMT promoter is closely associated with the survival period of glioma patients and their response to chemotherapy drug temozolomide. Predicting the promoter status of MGMT accurately pre-operator is crucial for making personalized treatment decisions for glioma patients.

Goal(s): To propose models based on CNNs to predict the MGMT methylation status of gliomas using conventional pre-operative MR images.

Approach: Building three CNNs models based on T2WI, T2-FLAIR, CE-T1WI images, respectively. Fusing features to build the fourth model to predict the MGMT methylation status.

Results: All models can predict the MGMT status effectively and accurately, the fused-feature model has the best diagnostic performance.

Impact: Models based on conventional MRI sequences and VASARI features provide the clinical value for evaluation of molecular typing in gliomas. It is expected to become a practical tool for the non-invasive characterization of gliomas to help the individualized treatment planning.

Keywords: Diagnosis/Prediction, Radiomics

Motivation: Radiomics has potential to bring added values to cancer diagnosis and prognosis, whilst a practical tool that is accessible for clinicians and radiologists has not been available.

Goal(s): To design MIROR, a clinical decision support system, that can aid tumour diagnosis with real-time image and spectroscopy analysis.

Approach: The project keeps collecting childhood brain tumour proton MR images and spectroscopy in England and has built a multi-centre database that includes 377 cases.

Results: MIROR supports key features including visualisation, image analysis, feature extraction, spectroscopy quantification, and tumour type and subtype prediction through machine learning.

Impact: A practical solution that translates radiomics and machine learning into clinical scenarios can aid tumour diagnosis and treatment planning, bring benefits to patient healthcare, and improve clinical outcomes.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Longitudinal health monitoring

Motivation: Techniques that allow automated evaluation of the evolution of disease risk over time can be of great value for active surveillance and other imaging-based monitoring. 

Goal(s): We introduce a novel self-supervised framework to learn representations that can identify increases in risk over time.

Approach: We propose a contrastive learning model to first learn subject-specific representations from low-slice-resolution images followed by learning a risk axis in the representational space to provide information on global changes in risk over time. 

Results: The developed framework was used to assess risk of new metastases in a cohort of subjects from the NYU-Mets longitudinal imaging dataset.

Impact: A key question when moving to lower field strengths in MRI is if we can get comparable information from lower-quality images as we can from the current standard of high-quality, high-resolution images. Self-supervised contrastive learning approaches can hold the key.