Abstract:Purpose: Brain metastases (BM) occur frequently in patients with metastatic cancer. Early and accurate detection of BM is essential for treatment planning and prognosis in radiation therapy. Due to their tiny sizes and relatively low contrast, small BM are very difficult to detect manually. With the recent development of deep learning technologies, several res earchers have reported promising results in automated brain metastasis detection. However, the detection sensitivity is still not high enough for tiny BM, and integration into clinical practice in regard to differentiating true metastases from false positives (FPs) is challenging. Methods: The DeepMedic network with the binary cross-entropy (BCE) loss is used as our baseline method. To improve brain metastasis detection performance, a custom detection loss called volume-level sensitivity-specificity (VSS) is proposed, which rates metastasis detection sensitivity and specificity at a (sub)volume level. As sensitivity and precision are always a trade-off, either a high sensitivity or a high precision can be achieved for brain metastasis detection by adjusting the weights in the VSS loss without decline in dice score coefficient for segmented metastases. To reduce metastasis-like structures being detected as FP metastases, a temporal prior volume is proposed as an additional input of DeepMedic. The modified network is called DeepMedic+ for distinction. Combining a high-sensitivity VSS loss and a high specificity loss for DeepMedic+, the majority of true positive metastases are confirmed with high specificity, while additional metastases candidates in each patient are marked with high sensitivity for detailed expert evaluation. Results: Our proposed VSS loss improves the sensitivity of brain metastasis detection, increasing the sensitivity from 85.3% for DeepMedic with BCE to 97.5% for DeepMedic with VSS. Alternatively, the precision is improved from 69.1% for DeepMedic with BCE to 98.7% for DeepMedic with VSS. Comparing DeepMedic+ with DeepMedic with the same VSS loss, 44.4% of the FP metastases are reduced in the high-sensitivity model and the precision reaches 99.6% for the high-specificity model. The mean dice coefficient for all metastases is about 0.81. With the ensemble of the high-sensitivity and high-specificity models, on average only 1.5 FP metastases per patient need further check, while the majority of true positive metastases are confirmed. Conclusions: Our proposed VSS loss and temporal prior improve brain metastasis detection sensitivity and precision. The ensemble learning is able to distinguish high confidence true positive metastases from metastases candidates that require special expert review or further follow-up, being particularly well-fit to the requirements of expert support in real clinical practice. This facilitates metastasis detection and segmentation for neuroradiologists in diagnostic and radiation oncologists in therapeutic clinical applications.

BrainLossNet: a fast, accurate and robust method to estimate brain volume loss from longitudinal MRI

Higher effect sizes for the detection of accelerated brain volume loss and disability progression in multiple sclerosis using deep-learning

Learning from pseudo-labels: deep networks improve consistency in longitudinal brain volume estimation

CerebNet: A fast and reliable deep-learning pipeline for detailed cerebellum sub-segmentation

MRI Volume-Based Robust Brain Age Estimation Using Weight-Shared Spatial Attention in 3D CNNs

Volume Preserving Brain Lesion Segmentation.

Convolutional Neural Network for Fully Automated Cerebellar Volumetry in Children in Comparison to Manual Segmentation and Developmental Trajectory of Cerebellar Volumes

DeepAtrophy: Teaching a Neural Network to Differentiate Progressive Changes from Noise on Longitudinal MRI in Alzheimer's Disease

Deepbet: Fast brain extraction of T1-weighted MRI using Convolutional Neural Networks

$ν$-net: Deep Learning for Generalized Biventricular Cardiac Mass and Function Parameters

Development and validation of a two-stage convolutional neural network algorithm for segmentation of MRI white matter hyperintensities for longitudinal studies in CADASIL

Toward hippocampal volume measures on ultra-high field magnetic resonance imaging: a comprehensive comparison study between deep learning and conventional approaches

Going deeper with brain morphometry using neural networks

DeepVolume: Brain Structure and Spatial Connection-Aware Network for Brain MRI Super-Resolution

Kernel Ridge Regression-based Randomized Network for Brain Age Classification and Estimation

CineVN: Variational network reconstruction for rapid functional cardiac cine MRI

Student beats the teacher: deep neural networks for lateral ventricles segmentation in brain MR

A Volumetric Convolutional Neural Network for Brain Tumor Segmentation

Accurate, rapid and reliable, fully automated MRI brainstem segmentation for application in multiple sclerosis and neurodegenerative diseases

Brain Age Estimation From MRI Using Cascade Networks with Ranking Loss

Deep learning for brain metastasis detection and segmentation in longitudinal MRI data