RESUMO
BACKGROUND: Differential diagnosis of brain metastases subtype and primary central nervous system lymphoma (PCNSL) is necessary for treatment decisions. The application of machine learning facilitates the classification of brain tumors, but prior investigations into primary lymphoma and brain metastases subtype classification have been limited. PURPOSE: To develop a machine-learning model to classify PCNSL, brain metastases with primary lung and non-lung origin. STUDY TYPE: Retrospective. POPULATION: A total of 211 subjects with pathologically confirmed PCNSL or brain metastases (training cohort 168 and testing cohort 43). FIELD STRENGTH/SEQUENCE: A 3.0 T axial contrast-enhanced T1-weighted spin-echo inversion recovery sequence (T1WI-CE), axial T2-weighted fluid-attenuation inversion recovery sequence (T2FLAIR) ASSESSMENT: Several machine-learning models (support vector machine, random forest, and K-nearest neighbors) were built with least absolute shrinkage and selection operator (LASSO) using features from T1WI-CE, T2FLAIR, and clinical. The model with the highest performance in the training cohort was selected to differentiate lesions in the testing cohort. Then, three radiologists conducted a two-round classification (with and without model reference) using images and clinical information from testing cohorts. STATISTICAL TESTS: Five-fold cross-validation was used for model evaluation and calibration. Model performance was assessed based on sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC). RESULTS: Twenty-five image features were selected by LASSO analysis. Random forest classifier was selected for its highest performance on the training set with an AUC of 0.73. After calibration, this model achieved an accuracy of 0.70 on the testing set. Accuracies of all three radiologists improved under model reference (0.49 vs. 0.70, 0.60 vs. 0.77, 0.58 vs. 0.72, respectively). DATA CONCLUSION: The random forest model based on conventional MRI and clinical data can diagnose PCNSL and brain metastases subtypes (lung and non-lung origin). Model classification can help foster the diagnostic accuracy of specialists and streamline prognostication workflow. EVIDENCE LEVEL: 4 TECHNICAL EFFICACY: Stage 2.
Assuntos
Neoplasias Encefálicas , Linfoma , Humanos , Estudos Retrospectivos , Neoplasias Encefálicas/patologia , Imageamento por Ressonância Magnética/métodos , Linfoma/diagnóstico por imagem , Linfoma/patologia , Sistema Nervoso Central/patologiaRESUMO
PURPOSE: We aimed to develop and validate a deep learning (DL) model to automatically segment posterior fossa ependymoma (PF-EPN) and predict its molecular subtypes [Group A (PFA) and Group B (PFB)] from preoperative MR images. EXPERIMENTAL DESIGN: We retrospectively identified 227 PF-EPNs (development and internal test sets) with available preoperative T2-weighted (T2w) MR images and molecular status to develop and test a 3D nnU-Net (referred to as T2-nnU-Net) for tumor segmentation and molecular subtype prediction. The network was externally tested using an external independent set [n = 40; subset-1 (n = 31) and subset-2 (n =9)] and prospectively enrolled cases [prospective validation set (n = 27)]. The Dice similarity coefficient was used to evaluate the segmentation performance. Receiver operating characteristic analysis for molecular subtype prediction was performed. RESULTS: For tumor segmentation, the T2-nnU-Net achieved a Dice score of 0.94 ± 0.02 in the internal test set. For molecular subtype prediction, the T2-nnU-Net achieved an AUC of 0.93 and accuracy of 0.89 in the internal test set, an AUC of 0.99 and accuracy of 0.93 in the external test set. In the prospective validation set, the model achieved an AUC of 0.93 and an accuracy of 0.89. The predictive performance of T2-nnU-Net was superior or comparable to that of demographic and multiple radiologic features (AUCs ranging from 0.87 to 0.95). CONCLUSIONS: A fully automated DL model was developed and validated to accurately segment PF-EPNs and predict molecular subtypes using only T2w MR images, which could help in clinical decision-making.
Assuntos
Aprendizado Profundo , Ependimoma , Humanos , Estudos Retrospectivos , Área Sob a Curva , Tomada de Decisão Clínica , Ácido Fenilfosfonotioico, 2-Etil 2-(4-Nitrofenil) Éster , Ependimoma/diagnóstico por imagem , Ependimoma/genética , Imageamento por Ressonância MagnéticaRESUMO
BACKGROUND AND PURPOSE: Symptoms of normal pressure hydrocephalus (NPH) are sometimes refractory to shunt placement, with limited ability to predict improvement for individual patients. We evaluated an MRI-based artificial intelligence method to predict post-shunt NPH symptom improvement. MATERIALS AND METHODS: NPH patients who underwent magnetic resonance imaging (MRI) prior to shunt placement at a single center (2014-2021) were identified. Twelve-month post-shunt improvement in modified Rankin Scale (mRS), incontinence, gait, and cognition were retrospectively abstracted from clinical documentation. 3D deep residual neural networks were built on skull stripped T2-weighted and fluid attenuated inversion recovery (FLAIR) images. Predictions based on both sequences were fused by additional network layers. Patients from 2014-2019 were used for parameter optimization, while those from 2020-2021 were used for testing. Models were validated on an external validation dataset from a second institution (n=33). RESULTS: Of 249 patients, n=201 and n=185 were included in the T2-based and FLAIR-based models according to imaging availability. The combination of T2-weighted and FLAIR sequences offered the best performance in mRS and gait improvement predictions relative to models trained on imaging acquired using only one sequence, with AUROC values of 0.7395 [0.5765-0.9024] for mRS and 0.8816 [0.8030-0.9602] for gait. For urinary incontinence and cognition, combined model performances on predicting outcomes were similar to FLAIR-only performance, with AUROC values of 0.7874 [0.6845-0.8903] and 0.7230 [0.5600-0.8859]. CONCLUSIONS: Application of a combined algorithm using both T2-weighted and FLAIR sequences offered the best image-based prediction of post-shunt symptom improvement, particularly for gait and overall function in terms of mRS. ABBREVIATIONS: NPH = normal pressure hydrocephalus; iNPH = idiopathic NPH; sNPH = secondary NPH; AI = artificial intelligence; ML = machine learning; CSF = cerebrospinal fluid; AUROC = area under the receiver operating characteristic; FLAIR = fluid attenuated inversion recovery; BMI = body mass index; CCI = Charlson Comorbidity Index; SD = standard deviation; IQR = interquartile range.
RESUMO
Uptake segmentation and classification on PSMA PET/CT are important for automating whole-body tumor burden determinations. We developed and evaluated an automated deep learning (DL)-based framework that segments and classifies uptake on PSMA PET/CT. We identified 193 [18F] DCFPyL PET/CT scans of patients with biochemically recurrent prostate cancer from two institutions, including 137 [18F] DCFPyL PET/CT scans for training and internally testing, and 56 scans from another institution for external testing. Two radiologists segmented and labelled foci as suspicious or non-suspicious for malignancy. A DL-based segmentation was developed with two independent CNNs. An anatomical prior guidance was applied to make the DL framework focus on PSMA-avid lesions. Segmentation performance was evaluated by Dice, IoU, precision, and recall. Classification model was constructed with multi-modal decision fusion framework evaluated by accuracy, AUC, F1 score, precision, and recall. Automatic segmentation of suspicious lesions was improved under prior guidance, with mean Dice, IoU, precision, and recall of 0.700, 0.566, 0.809, and 0.660 on the internal test set and 0.680, 0.548, 0.749, and 0.740 on the external test set. Our multi-modal decision fusion framework outperformed single-modal and multi-modal CNNs with accuracy, AUC, F1 score, precision, and recall of 0.764, 0.863, 0.844, 0.841, and 0.847 in distinguishing suspicious and non-suspicious foci on the internal test set and 0.796, 0.851, 0.865, 0.814, and 0.923 on the external test set. DL-based lesion segmentation on PSMA PET is facilitated through our anatomical prior guidance strategy. Our classification framework differentiates suspicious foci from those not suspicious for cancer with good accuracy.