Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: There is currently no reliable tool to predict postoperative recurrence for patients who undergo surgery for brain metastases (BrMs).

Goal(s): This study aimed to develop and externally validate a prognostic model to predict intracranial recurrence and recurrence-free survival (RFS) for lung cancer patients receiving BrM surgery. 

Approach: A combined prognostic model-based nomogram was developed by incorporating clinical and structural MRI predictors, radiomics and deep signatures extracted from MR images.

Results: The nomogram predicted accurately for RFS and intracranial recurrence prediction, both in the training and test sets .

Impact: The combined prognostic model-based nomogram can be used as a preoperative tool to predict intracranial recurrence and recurrence-free survival after surgical resection of brain metastases in lung cancer patients.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: In clinical practice, distinguishing atypical radiological features of primary central nervous system lymphoma (PCNSL) from certain glioma or demyelinating disease patients is challenging and often lead to delayed or incorrect treatment.

Goal(s): To develop deep learning model to identify PCNSL with atypical radiological features.

Approach: Developing a multi-task, multi-modal deep learning model capable of end-to-end identification of early-stage atypical PCNSL.

Results: The multi-task, multi-modal deep learning model accurately discriminates early-stage atypical PCNSL from other radiologically similar diseases, significantly enhancing the diagnostic accuracy of radiologists in clinical practice.

Impact: The practical clinical application of the model demonstrates its diagnostic value in identifying challenging cases suspected of early-stage atypical PCNSL. This research shifts academic attention towards distinguishing specific subtypes prone to misdiagnosis, rather than solely focusing on disease-level differentiations.

Keywords: Diagnosis/Prediction, Brain

Motivation: Infiltration and recurrence of glioblastoma is typically fatal. Conventional imaging techniques are insufficient for identifying the infiltrated regions.

Goal(s): We aim to develop an interactive visualization method based on conventional MRI to identify the peri-tumor infiltration.

Approach: Glioblastoma infiltrating area detection interactive framework (GIADIF) consists of two steps: delineating peritumoral edema and extracting the voxels with low fractional anisotropy value as user-interactive input; using the P-Net from the DeepIGeoS framework to output the infiltrated maps, and validating in a prospective cohort.

Results: GIADIF showed reliable performance in identifying GBM-infiltrated regions (area under the receiver operating characteristic curve: 0.929 [95% CI 0.804–1.000]).

Impact: GIADIF utilizes the interactive information to the conventional MRI sequences to locate areas of GBM infiltration. Its excellent performance allows for the prompt and precise selection sites for surgery and radiotherapy.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, brain metastasis

Motivation: Lung cancer is the most common primary tumor showing brain metastasis (BM). Epidermal growth factor receptor (EGFR) mutations are detected in a significant proportion of lung cancer patients. 

Goal(s): However, a subset of patients may show discordance in EGFR mutation status between the primary lung tumor and the corresponding BMs, which may affect decision-making in treatments.

Approach: We used machine learning (ML) based on pretreatment brain MRI and clinical data for prediction of  EGFR mutation status in BMs of lung cancer.

Results: Among various ML algorithms, the best predictive performance with accuracy of 89%, precision of 88%, and AUC of 0.97 were obtained.

Impact: Machine learning based on pretreatment clinical data and brain MRI provides the potential to predict the EGFR mutation status in brain metastasis of lung cancer, and may affect decision-making in treatments.

Keywords: Diagnosis/Prediction, Brain

Motivation: Genetic biomarkers and WHO grading of gliomas are critical for the classification of glioma subtypes, treatment planning and survival prognosis.  

Goal(s): The aim of this study is to apply DL network for non-invasive prediction of multiple genes and classification of subtypes.

Approach: A decision tree diagnostic scheme based on multi-label DL network was constructed to classify adult-type diffuse gliomas into 5 subtypes based on the 2021 WHO classification of tumor of the CNS, combining the WHO grading and 3 genetic biomarkers status.

Results: The model we developed can reclassify adult-type diffuse glioma with a diagnostic accuracy of 94.4%.

Impact: Based on the 2021 WHO CNS tumor classification, this study applies multi-label deep learning to reclassify adult-type diffuse gliomas, which can be helpful for patients to obtain preoperative diagnosis and precise treatment.

Keywords: Diagnosis/Prediction, Brain, Radiomics; Meningioma; Machine Learning; Hemorrhage; Cerebral Edema

Motivation: Prediction radiomics analysis of postoperative progressive cerebral edema and hemorrhage which are the most common complications after meningioma resection, is limited.

Goal(s): To develop and validate a machine learning model to predict progressive cerebral edema and hemorrhage after meningioma resection.

Approach: Reviewing the preoperative MRI of 148 pathology-confirmed meningiomas, extracting radiomics features of tumor enhancement and peritumoral edema regions, and combining clinical characteristics to build machine learning multiparametric MRI radiomics predictive models.

Results: The combining model including both enhancement and edema radiomics features, and clinical characteristics including systolic blood pressure, showed the best predictive performance with AUC of 0.94 for the validation set.

Impact: We proposed a novel model that included clinical indicators and multi-parameter radiomics features, which can accurately and non-invasively predict progressive cerebral edema and hemorrhage after meningioma resection, enabling improving clinical management and quality of life of patients with meningioma.

Keywords: Diagnosis/Prediction, Cancer, glioma

Motivation: Early detection of low-grade glioma (LGG) malignant transformation (MT) is vital for treatment decisions, prognosis, quality of life and patient-centered care.

Goal(s): To develop non-linear machine learning models using XGBoost algorithm to predict overall survival using clinical, molecular, genetic and radiomic data at MT. 

Approach: 553 LGGs with histology and MRI underwent in-house tumour segmentation pipeline with radiomic feature extraction and masked disconnectome of map components.

Results: XGB Classifier model predicted OS > 5 years from MT with an accuracy of 64%. Age, IDH1 mutation, 1p/19q co-deletion, regularity of tumour shape, and disconnectome-related perilesional components were most predictive of survival.

Impact: Understanding malignant transformation of low-grade gliomas is crucial for research and the development of new treatment strategies. Defining the radiological features at malignant transformation allows for a timely shift in the treatment plan with potential to improve repsonse to therapy.

Keywords: Diagnosis/Prediction, Radiomics, Gliomas

Motivation: The IDH1 mutant state is an independent risk factor of affecting the treatment and prognosis of glioma. Predicting the IDH1 status accurately pre-operator is crucial for making personalized treatment decisions for glioma patients.

Goal(s): This study aims to propose a non-invasive and convenient model based on MRI to predict the IDH1 status of gliomas before operation accurately.

Approach: Building three machine learning models based on multi-sequence MRI radiomics features, VASARI features, and combined features to predict the IDH1 status.

Results: These three models can predict the IDH1 status effectively and accurately, the combined 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, bone invasion,meningiomas

Motivation: Bone invasion is a common problem in meningioma surgery and is associated with patient prognosis. However, 10-26% of patients with potential bone invasion are difficult to identify by preoperative imaging.

Goal(s): To develop an artificial intelligence based preoperative diagnostic model.

Approach: Radiomics features were extracted from preoperative, contrast-enhanced T1-weighted (T1C) and T2-weighted (T2) MR images of 296 patients. Candidate radiomics were selected by applying feature reduction and 5-fold cross validation.

Results: A more accurate and robust fusion radiomics model was built based on T1C and T2 MR images with AUC of 0.755.

Impact: Our results have demonstrated that radiomics features extracted from T1C and T2 MR images may be employed as effective preoperative biomarkers for predicting potential bone invasion in meningiomas.

Keywords: Diagnosis/Prediction, Tumor, Glioblastomas, Solitary_Brain_Metastases, Quantitative_Physiologic, MRI, Innovative_Hyper-tuned_Machine_Learning

Motivation: There is an unmet need to develop advanced MRI based prediction models to distinguish glioblastomas (GBMs) from solitary brain metastases (BMs) with high-accuracy, as conventional MRI-techniques often yield ambiguous results.

Goal(s): The objective is to discriminate GBMs from solitary BMs using physiologic MRI parameters and machine-learning based novel-methods.

Approach: Employing diffusion-tensor-imaging (DTI) and dynamic-susceptibility contrast-perfusion weighted imaging (DSC-PWI), the study uses a novel machine-learning approach that integrates multiple features with hyper-tuned models to enhance pattern-recognition and prediction.

Results: The innovative-method combining interacted and non-interacted features via hyper-tuned machine-learning models significantly outperformed traditional-methods, thus achieving high accuracy and reliability in differentiating GBMs from BMs.

Impact: The integration of quantitative and physiologically-sensitive MRI-parameters with novel machine-learning based algorithms may be promising in distinguishing glioblastomas from solitary brain metastases. This approach may be useful in making prognostication and guiding optimal, personalized patient-treatments in the era of personalized-medicine.

Keywords: Diagnosis/Prediction, Radiomics

Motivation: Investigated the underlying impact of subregional analysis on model performance: comparison of two volumes of interests (VOI) definition strategies.

Goal(s): To explore a subregion-based RadioFusionOmics (RFO) model for discrimination between adult-type grade 4 astrocytoma and glioblastoma.

Approach: Subregional radiomics analysis using the K-means clustering demonstrated discriminative performance comparable to that of manual segmentation. Edematous subregion is a possible intratumoral heterogeneity phenotype that differentiates grade 4 astrocytoma from glioblastoma.

Results: The RFO model that was trained using fused features  achieved the AUC of 0.868 (VOI₃) and 0.884 (H₃₄) in the primary cohort (p=0.059), and 0.824 (VOI₃) and 0.838 (H₃₄) in the testing cohort (p=0.023).

Impact: Fusion of features from edematous subregions of multiple MRI sequences by the RFO model identified IDH genotypes of adult type grade 4 gliomas in line with current WHO CNS 5 criteria.

Keywords: Diagnosis/Prediction, Radiomics, C5aR1; high-grade glioma; MRI; prognosis; biomarker

Motivation: High-grade glioma is a complex disease characterized by genome instability caused by the accumulation of genetic alterations. Identifying and evaluating the oncogenes involved is crucial for determining treatment strategies and evaluating prognosis.

Goal(s): We sought to explore whether radiomics models based on MRI features can noninvasively predict C5aR1 expression and the prognosis of patients with high-grade glioma.

Approach: This study uses machine learning approaches based on paired MRI and RNA sequencing data.

Results: The radiomics models yield satisfactory performances in predicting C5aR1 expression. Our findings also reveal associations between MRI radiomics and immune-related features.

Impact:  As an effective and reproducible tool, our radiomics model may support clinical decision making and individualized treatment.

Keywords: Diagnosis/Prediction, Radiomics, glioma, deep learning radiomics nomograms

Motivation: It is unclear whether deep learning radiomics nomograms (DLRN) can noninvasively predict isocitrate dehydrogenase (IDH) genotypes in glioma patients.
Goal To explore the feasibility of DLRN in predicting IDH genotype.

Goal(s): To explore the feasibility of DLRN in predicting IDH genotype.

Approach: T2WI-based DLRN was developed and validated in two centers (Center I, n=342 and Center II, n=60) to predict IDH genotype and evaluate its association with prognosis in glioma patients.

Results: The proposed model had an area under the curve(AUC)of 0.98 in an externally validated cohort, and DLRN scores were significantly associated with the overall survival of glioma patients.

Impact: The proposed DLRN can accurately predict IDH genotypes and provide a useful tool for targeted therapy of patients with IDH mutations.

Keywords: Diagnosis/Prediction, Radiomics

Motivation: Magnetic resonance imaging (MRI) raidomics has shown unique advantages and potential in non-invasive evaluation of therapeutic efficacy in cancer patients.

Goal(s): To construct a model that can predict the prognosis of patients with advanced non-small cell lung cancer (NSCLC) brain metastases after radiotherapy.

Approach: For patients with advanced NSCLC brain metastasis who underwent pre-treatment MRI examination, stable and reproducible raidomics features were quantitatively extracted and screened. Additionally, artificial intelligence methods were utilized to construct raidomics labels.

Results: The potential of the MRI raidomics-based method in predicting the efficacy of radiotherapy for brain metastases from advanced NSCLC was preliminarily confirmed.

Impact: The method based on MRI raidomics (T1, T2, DWI, DCE) can not only enhance the precision of radiation therapy efficacy assessment for patients with NSCLC brain metastasis, but also offer clinicians a more scientific basis for treatment decision-making.

Keywords: AI/ML Software, Brain, Deep Learning, Attention

Motivation: There is a need to preoperatively assess the isocitrate dehydrogenase (IDH) mutational status in gliomas, which highly affects the treatment planning and patient prognosis.

Goal(s): To develop a robust deep learning pipeline for noninvasively assessing the IDH mutational status of gliomas based on anatomical MRI

Approach: Post-contrast T1-weighted, T2-weighted, and fluid-attenuated inversion recovery (FLAIR) MRI of 501 adult diffuse gliomas (103 IDH-mutant, 308 IDH-wildtype) of the UCSF-PGDM dataset were evaluated with a 2D UNet architecture using synthetic attention.

Results: The model utilizing all three anatomical modalities achieved an accuracy of 93.31% (sensitivity=93.33%, specificity=93.24%). 

Impact: IDH mutational status in gliomas was identified with over 93% accuracy utilizing a 2D UNet architecture with synthetic attention for the evaluation of whole tumor slices of three standard anatomical MRI modalities. 

Keywords: Diagnosis/Prediction, Tumor, Neuro-oncology

Motivation: Accurate glioma classification currently relies on tissue diagnosis, which has associated surgical risks. Machine learning based classification of MR images may enable non-invasive glioma characterisation.

Goal(s): Our aim was to assess which imaging modalities provided optimal training data to increase accuracy of machine learning based glioma characterisation.

Approach: A pyRadiomics based pipeline predicted tumour grade and IDH-mutation status with XGBoost on a glioblastoma-rich dataset. 10 structural and advanced MR acquisitions were used as model input and a systematic search for the most informative MR modalities was performed.

Results: The classifier performed best when the model was trained on post-contrast T1 and diffusion imaging.

Impact: We found post-contrast T1 and diffusion imaging to be the most informative MR modalities for machine learning based glioma characterisation. This result will benefit scientists in making well-informed choices on how to train their machine learning models for glioma classification.