Bsci-Seg: A Bone Scintigram Segmentation Model for Accurately Detecting and Delineating Metastasis Lesions Using Convolutional Neural Networks

Abstract

Lesion segmentation is a crucial step in the automated analysis of bone scintigrams. The rapid development of convolutional neural networks provides a novel approach for bone metastasis lesion segmentation. Previous attempts have been made to enhance the performance of relevant algorithms by improving U-Net based models. However, the interpretability and lesion detection rates of these methods require further enhancement. This paper introduces a novel bone scintigram segmentation model named BSci-Seg, which is inspired by diagnostic patterns and aims to tackle the aforementioned issues. BSci-Seg is structured upon a classical encoding-decoding architecture and incorporates a dual sampling scheme, a multiple receptive-field attention, and a customized loss function. The segmentation network employs a dual sampling scheme in both stages, incorporating multiple receptive field attentions during the feature encoding phase to improve the model’s capacity to detect occult lesions. Experimental evaluations conducted on 286 SPECT bone scintigrams show remarkable increase in both Dice Similarity Coefficient (DSC) and Recall scores, by 4.53% and 9.90%, respectively. Additionally, a comparison with existing models for bone metastasis lesion segmentation validates the superior performance of our proposed method. Comprehensive ablation studies and case studies highlight the meaningful implications, providing a foundation for further in-depth research.

Note:
Funding declaration: This work was supported in part by the National Natural Science Foundation of China [grant numbers 62362058, 61562075], the Gansu Province’s Key Provincial Talents Program, the Young and Middle-aged Talents Training Program of State Ethnic Affairs Commission, the Key R&D Plan of Gansu Province [grant number 21YF5GA063], the Natural Science Foundation of Gansu Province under [grant numbers 20JR5RA511, 22JR11RA236], the Fundamental Research Fund for the Central Universities under Grants [grant numbers 31920230172, 31920220020, 31920220054], and the Youth Ph.D. Foundation of Education Department of Gansu Province [grant number 2021QB-063]

Conflict of Interests: None.

Ethics statement: The study was approved by the Ethics Committee of Gansu Provincial Tumor Hospital (Approval No.: A202106100014).

Keywords: Tumor bone metastasis, Bone scintigram, Lesion segmentation, Knowledge incorporation, Convolutional Neural Network