Automated Brain Metastases Detection Framework for T1-Weighted Contrast-Enhanced 3D MRI

• Published in: IEEE journal of biomedical and health informatics Vol. 24; no. 10; pp. 2883 - 2893
• Main Authors: Dikici, Engin, Ryu, John L., Demirer, Mutlu, Bigelow, Matthew, White, Richard D., Slone, Wayne, Erdal, Barbaros Selnur, Prevedello, Luciano M.
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.10.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Brain; Brain - diagnostic imaging; Brain cancer; brain metastases; Brain Neoplasms - diagnostic imaging; Brain Neoplasms - secondary; Comparative analysis; computer-aided detection; convolutional neural networks; Datasets; Deep Learning; Elastic deformation; Gadolinium; Humans; Image Interpretation, Computer-Assisted - methods; Imaging, Three-Dimensional - methods; Informatics; Lesions; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; medical image analysis; Metastases; Metastasis; Neural networks; Neural Networks, Computer; Neuroimaging; scale-space representations; Sensitivity; Sensitivity and Specificity; Strain; Three-dimensional displays; Training
• ISSN: 2168-2194; 2168-2208
• DOI: 10.1109/JBHI.2020.2982103

Brain Metastases (BM) complicate 20-40% of cancer cases. BM lesions can present as punctate (1 mm) foci, requiring high-precision Magnetic Resonance Imaging (MRI) in order to prevent inadequate or delayed BM treatment. However, BM lesion detection remains challenging partly due to their structural similarities to normal structures (e.g., vasculature). We propose a BM-detection framework using a single-sequence gadolinium-enhanced T1-weighted 3D MRI dataset. The framework focuses on the detection of smaller (<15 mm) BM lesions and consists of: (1) candidate-selection stage, using Laplacian of Gaussian approach for highlighting parts of an MRI volume holding higher BM occurrence probabilities, and (2) detection stage that iteratively processes cropped region-of-interest volumes centered by candidates using a custom-built 3D convolutional neural network ("CropNet"). Data is augmented extensively during training via a pipeline consisting of random ga mma correction and elastic deformation stages; the framework thereby maintains its invariance for a plausible range of BM shape and intensity representations. This approach is tested using five-fold cross-validation on 217 datasets from 158 patients, with training and testing groups randomized per patient to eliminate learning bias. The BM database included lesions with a mean diameter of ∼5.4 mm and a mean volume of ∼160 mm 3 . For 90% BM-detection sensitivity, the framework produced on average 9.12 false-positive BM detections per patient (standard deviation of 3.49); for 85% sensitivity, the average number of false-positives declined to 5.85. Comparative analysis showed that the framework produces comparable BM-detection accuracy with the state-of-art approaches validated for significantly larger lesions.