Whole-Body Bone Scan Dataset (BS-80K)

This repository contains information about the dataset and methodology used in the study of whole-body bone scans for identifying metabolic abnormalities. The dataset originates from BS-80K, a collection of bone scan images curated at West China Hospital.

Dataset Description

The dataset used in this study is derived from BS-80K and contains a total of 3,253 whole-body bone scan images, categorized into the following:

• Anterior images: 2,924 images with bounding box annotations.
• Posterior images: 2,555 images with bounding box annotations.

Classes

The dataset consists of three distinct classes:

1. Background
2. Abnormal (regions indicative of abnormal metabolic activity)
3. Normal

Image Specifications

• Resolution: Each image is 256 x 1024 pixels.
• Annotations: Bounding box annotations are provided for precise segmentation and accurate identification of areas of interest.

Importance of Bounding Box Annotations

The bounding box annotations in this dataset are essential for:

1. Precise Segmentation: Helping delineate regions of interest for further analysis.
2. Accurate Identification: Supporting the detection of abnormal metabolic activity in whole-body bone scans.

Reference Study

This dataset and methodology are part of a study conducted at West China Hospital. You can find more details in the original paper:

• Paper Title: Deep Learning-Assisted Detection of Metabolic Abnormalities in Whole-Body Bone Scans
• Published On: PubMed

Usage

Researchers and practitioners can use this dataset for tasks such as:

• Image Classification
• Object Detection
• Medical Image Segmentation

If you utilize this dataset, please make sure to cite the original study appropriately.

---

Citation

Please cite the paper if you use this dataset in your research:

[13] Deep Learning-Assisted Detection of Metabolic Abnormalities in Whole-Body Bone Scans, PubMed ID: 36334360.

U-Net++

UNet++ is an architecture for semantic segmentation based on the U-Net. Through the use of densely connected nested decoder sub-networks, it enhances extracted feature processing and was reported by its authors to outperform the U-Net in Electron Microscopy (EM), Cell, Nuclei, Brain Tumor, Liver and Lung Nodule medical image segmentation tasks.

Key Features

• Nested Dense Skip Pathways: Redesigned skip connections reduce the semantic gap between encoder and decoder feature maps, enabling the optimizer to learn more efficiently.
• Deep Supervision: Multi-level supervision during training improves convergence and segmentation accuracy.

Reference

• Paper: UNet++: A Nested U-Net Architecture for Medical Image Segmentation
• Authors: Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, Jianming Liang
• Published: Accepted by the 4th Deep Learning in Medical Image Analysis (DLMIA) Workshop.

Citation

If you find this work useful, please cite the original paper

Paper Under Review

Currently, I am preparing a paper based on this dataset and the methodology used in this research. The paper is under review and will be published soon. I kindly ask that this code and dataset not be used, shared, or modified until the paper is officially published. Thank you for your understanding.

This code is proprietary and is shared for reference purposes only. It may not be used, modified, or distributed without explicit permission from the author. Use of this code is restricted until the related research paper is officially published