Tracking temporal progression of benign bone tumors through X-ray based detection and segmentation

Posted by mhb on 2025-11-16 10:07:31 |

Share: Facebook | Twitter | Whatsapp | Linkedin Visits: 82

Introduction

Benign bone tumors are commonly diagnosed using X-ray imaging because the method is cost-effective, accessible, and reliable. Physicians examine bone structure, transition zones, and lesion borders to distinguish benign tumors from aggressive ones. A narrow transition zone with clear margins usually indicates a benign bone tumor, while a permeative or unclear border may suggest a more aggressive condition. Monitoring these tumors across multiple time points is essential because changes in size, shape, and internal features help guide treatment decisions and determine whether further tests such as MRI or bone scans are necessary. However, traditional comparisons rely on visual judgment, leading to inconsistent interpretations among clinicians.

Recent advancements in artificial intelligence have encouraged the development of automated tools to support bone tumor analysis. These tools aim to reduce subjective variability and enhance the accuracy of detecting, segmenting, and monitoring benign bone tumors over time.

Related Work

Previous AI models have contributed valuable progress in bone tumor analysis. Earlier research introduced deep learning systems capable of segmenting and classifying tumors on radiographs. Although these models achieved strong accuracy, they lacked the ability to quantify tumor dimensions or track changes between different X-ray sessions. Other approaches focused on detecting specific tumor types but did not provide real-world measurements such as tumor length, width, or area. These limitations highlight the need for a comprehensive framework that supports both precise segmentation and longitudinal analysis.

Proposed Method: FusionX-BBTNet

FusionX-BBTNet is a newly developed AI framework designed to detect benign bone tumors, perform high-precision segmentation, calculate true tumor size, and compare sequential X-ray images. The model uses wavelet-based image enhancement to improve feature extraction and optical character recognition to convert pixel distances into millimeters using the scale bar printed on X-ray films. This combination allows accurate tumor measurement and objective monitoring of tumor progression.

Enhanced Image Construction

The method applies wavelet transforms to create enriched images containing detailed structural, textural, and boundary information. These enhanced components are merged with original grayscale images, enabling the deep learning model to learn richer features and improve segmentation performance.

Tumor Segmentation

The framework trains a U-Net model using a specially constructed dataset that includes both original and wavelet-enhanced images. This approach significantly improves segmentation accuracy, ensuring reliable identification of tumor boundaries and shapes. The model’s consistent performance demonstrates its suitability for clinical applications.

Quantitative Tumor Measurement

FusionX-BBTNet automatically extracts the X-ray scale bar and converts pixel lengths into real-world measurements. It calculates tumor width, height, and area using bounding box dimensions and pixel counts. These values are displayed directly on the output images, providing clear, interpretable results for clinical review. This process solves the common challenge of inconsistent or approximate tumor measurements in radiographic analysis.

Evaluation and Results

Performance on Internal Dataset

The proposed model achieved high accuracy in detection and segmentation tasks, demonstrating excellent reliability in identifying benign bone tumors. Key performance indicators include a high mean accuracy, strong intersection-over-union scores, and outstanding detection precision using YOLO-based methods.

Generalization on Public Dataset

FusionX-BBTNet also performed effectively on an external public dataset, achieving strong detection and segmentation scores without retraining. Although tumor size could not be calculated due to the absence of scale bars in the dataset, the model still demonstrated robust adaptability across different imaging conditions.

Conclusion

FusionX-BBTNet provides a powerful and reliable AI solution for analyzing benign bone tumors on X-ray images. By integrating advanced image fusion, deep learning segmentation, and automated scale-bar measurement, the framework enables accurate tumor detection and objective monitoring over time. This system reduces subjective variability, supports clinical decision-making, and offers a strong foundation for future enhancements using more advanced fusion and transformer-based architectures. Its high performance on internal and external datasets highlights its potential to become a valuable tool in orthopedic imaging and medical AI research.

Read Full Paper Here