BM-Seg: A new bone metastases segmentation dataset and ensemble of CNN-based segmentation approach

Abstract

In recent years, Machine Learning approaches (ML) have shown promising results in addressing many tasks in medical image analysis. In particular, the analysis of Bone Metastases (BM) has attracted considerable interest from both the medical and computer vision communities due to its critical and challenging aspect. Despite the research efforts, the detection of BM is still an open problem, mainly due to the lack of available datasets. This is due to two main obstacles: (i) the enormous time required for data collection and annotation, and (ii) privacy constraints. To overcome these challenges, we propose BM-Seg, a new dataset for segmenting BM from CT-scans. Our BM-Seg dataset consists of 1517 CT images from 23 patients where BM and bone regions were labeled by three radiologists. BM-Seg is constructed to cover the diversity of bone metastases in terms of location, organ and severity.We also propose a new CNN-based approach to segmentation of BM, presenting two main contributions. First, we introduce Hybrid-AttUnet++, a new Unet++ derived architecture with dual decoders that performs segmentation of BM and bone regions simultaneously. Second, we use an ensemble of trained Hybrid-AttUnet++ models (EH-AttUnet++) to optimize segmentation performance. Our experiments show that the EH-AttUnet++ architecture achieves better performance compared to state-of-the-art approaches for various evaluation metrics. The purpose of this work is to provide a benchmark dataset with new state-of-the-art performance in bone metastasis segmentation. This will facilitate further research in this area and help to put automatic detection and segmentation of bone metastases into practice.