PURPOSE: To demonstrate automatic detection of BM on three MRI datasets using a deep learning-based approach. To improve the performance of the network is iteratively co-trained with datasets from different domains. A systematic approach is proposed to prevent catastrophic forgetting during co-training.
STUDY TYPE: Retrospective.
POPULATION: A total of 156 patients (105 ground truth and 51 pseudo labels) with 1502 BM (BrainMetShare); 121 patients with 722 BM (local); 400 patients with 447 primary gliomas (BrATS). Training/pseudo labels/validation data were distributed 84/51/21 (BrainMetShare). Training/validation data were split: 121/23 (local) and 375/25 (BrATS).
FIELD STRENGTH/SEQUENCE: A 5 T and 3 T/T1 spin-echo postcontrast (T1-gradient echo) (BrainMetShare), 3 T/T1 magnetization prepared rapid acquisition gradient echo postcontrast (T1-MPRAGE) (local), 0.5 T, 1 T, and 1.16 T/T1-weighted-fluid-attenuated inversion recovery (T1-FLAIR) (BrATS).
ASSESSMENT: The ground truth was manually segmented by two (BrainMetShare) and four (BrATS) radiologists and manually annotated by one (local) radiologist. Confidence and volume based domain adaptation (CAVEAT) method of co-training the three datasets on a 3D nonlocal convolutional neural network (CNN) architecture was implemented to detect BM.
STATISTICAL TESTS: The performance was evaluated using sensitivity and false positive rates per patient (FP/patient) and free receiver operating characteristic (FROC) analysis at seven predefined (1/8, 1/4, 1/2, 1, 2, 4, and 8) FPs per scan.
RESULTS: The sensitivity and FP/patient from a held-out set registered 0.811 at 2.952 FP/patient (BrainMetShare), 0.74 at 3.130 (local), and 0.723 at 2.240 (BrATS) using the CAVEAT approach with lesions as small as 1 mm being detected.
DATA CONCLUSION: Improved sensitivities at lower FP can be achieved by co-training datasets via the CAVEAT paradigm to address the problem of data sparsity.
LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY STAGE: 2.
MATERIALS AND METHOD: This work uses two (private and public) datasets. The private dataset consists of 3807 magnetic resonance imaging (MRI) and computer tomography (CT) images belonging to two (normal and AD) classes. The second public (Kaggle AD) dataset contains 6400 MR images. The presented classification model comprises three fundamental phases: feature extraction using an exemplar hybrid feature extractor, neighborhood component analysis-based feature selection, and classification utilizing eight different classifiers. The novelty of this model is feature extraction. Vision transformers inspire this phase, and hence 16 exemplars are generated. Histogram-oriented gradients (HOG), local binary pattern (LBP) and local phase quantization (LPQ) feature extraction functions have been applied to each exemplar/patch and raw brain image. Finally, the created features are merged, and the best features are selected using neighborhood component analysis (NCA). These features are fed to eight classifiers to obtain highest classification performance using our proposed method. The presented image classification model uses exemplar histogram-based features; hence, it is called ExHiF.
RESULTS: We have developed the ExHiF model with a ten-fold cross-validation strategy using two (private and public) datasets with shallow classifiers. We have obtained 100% classification accuracy using cubic support vector machine (CSVM) and fine k nearest neighbor (FkNN) classifiers for both datasets.
CONCLUSIONS: Our developed model is ready to be validated with more datasets and has the potential to be employed in mental hospitals to assist neurologists in confirming their manual screening of AD using MRI/CT images.