An Intelligent Hybrid System for Fracture Detection and Healing Time Estimation from X- ray Images Using Lead-Lag CompensatorAnalyser

M. Sridevi, Svetha Venkatesh, P. Madhavasarma, P. Veeraragavan, Sylvain RavelResearch Square 2025 · DOI

TL;DR

Researchers have developed a hybrid system that uses a single X-ray image to both detect bone fractures and predict healing time with high accuracy. By applying control-theoretic principles—specifically transfer functions and lead-lag compensators—to image features, the system can estimate recovery periods within a 2–3% error margin, potentially eliminating the need for repeated radiation exposure from serial radiographs.

What Problem Does This Paper Address?

Monitoring the progress of a bone fracture typically involves "serial radiography"—taking multiple X-rays over weeks or months to see how the bone is knitting back together. This approach has three major downsides: cumulative radiation exposure for the patient, increased healthcare costs, and a reliance on subjective manual assessment by radiologists.

The authors sought to answer a challenging question: Can we look at the first diagnostic X-ray and predict the entire healing trajectory? To do this, they needed to move beyond simple image classification and find a way to model the biological "dynamics" of bone repair using static image data.

Methods

The authors propose a multi-stage "Intelligent Hybrid System" that blends computer vision, machine learning (ML), and classical control engineering.

  1. Image Processing & Feature Extraction: After standardizing X-ray images (grayscale conversion and Canny edge detection), the system extracts geometric features: fracture area, perimeter, and centroid coordinates ($X, Y$).
  2. Fracture Detection (The ML Stage): These features are fed into supervised learning models. The researchers tested three: Coarse Trees, Ensemble Bagged Trees (EBT), and Linear Support Vector Machines (LSVM).
  3. Healing Estimation (The Control Theory Stage): This is the paper's most novel contribution. The authors treat the fractured bone as an "unstable dynamic system" that transitions toward a "stable state" (healing). They convert the image's centroid data into a transfer function—a mathematical representation of the system's behavior.
  4. Lead-Lag Compensation: In control engineering, a lead-lag compensator is used to improve a system's stability and response time. Here, the authors designed a compensator to reduce "phase lag" in their mathematical model. By minimizing this lag, they significantly improved the precision of the healing time prediction formula.

Datasets

The study utilized a dataset of over 1,000 X-ray images of tibial (shinbone) fractures acquired from the Thanjavur Medical College Hospital database. The dataset included 21 distinct clinical cases, following patients from the initial fracture through to complete union (healing). The researchers used these real-world recovery timelines to validate their model's predictions.

Key Insights and Ideas

  • Bone as a Dynamic System: The core "aha moment" is the realization that bone healing kinetics can be modeled using stability principles. A fracture is a state of mechanical instability; healing is the process of returning to equilibrium.
  • Phase Lag as a Biomarker: The authors identified "phase lag" (derived from the transfer function of the image features) as a primary indicator of healing progress. A longer phase lag correlates with a more complex fracture and a longer recovery time.
  • Control Theory in Medicine: While lead-lag compensators are usually found in DC motor controllers or industrial robotics, this paper demonstrates they can effectively "tune" medical predictions by accounting for the distortions and delays inherent in biological data.

Key Results

  • Detection Accuracy: The Coarse Tree model outperformed others, achieving 90% accuracy in identifying whether a fracture was present. The LSVM and EBT models followed at 80% and 70%, respectively.
  • Prediction Precision: For the clinical case studies presented, the system’s predicted healing times were remarkably close to actual clinical outcomes. For example, in "Case 1," the actual healing time was 70 days; the system predicted 72 days—a marginal error of approximately 2.8%.
  • Voltage Sensitivity: The model incorporates a "user-input voltage" (simulating electrical stimulation therapy). The results suggested that a stimulation of 0.45V provided the most stable and optimal healing predictions.

Limitations and Future Work

While promising, the authors acknowledge several hurdles:

  • Image Quality Sensitivity: The system’s accuracy is highly dependent on the quality of the input X-ray and the consistency of pre-processing.
  • Generalization: The current study focused specifically on tibia fractures. Different bones (like the femur or radius) have different blood supply levels and mechanical loading environments, which might require recalibrating the transfer functions.
  • Biological Nuance: The model uses "generalized" transfer functions. Future work aims to integrate more patient-specific variables—such as age, bone density, or comorbidities like diabetes—and potentially incorporate real-time feedback from wearable biomedical sensors to further personalize the recovery roadmap.

Summary updated

Abstract

<title>Abstract</title> The traditional monitoring of bone fracture healing relies on multiple X-ray images, which can expose patients to unnecessary radiation. To address this, this work introduces a new method based on a transfer function derived from X-ray images to estimate bone healing time, reducing the need for repeated imaging. Data collected from Thanjavur Medical College Hospital initially identifies whether a fracture is present. The healing time is then determined using a specific approach. The method focuses on tibia fractures and justifies using a single original X-ray image to extract key fracture-related features such as phase lag, fracture area, fracture perimeter, and centroid coordinates (X, Y). These features train three machine learning models—Coarse Tree, Ensemble Bagged Trees, and Linear SVM—to classify fractures, achieving accuracies of 90%, 70%, and 80%, respectively. The centroid data used to estimate healing time are converted into a transfer function that describes the fracture condition. The phase lag derived from this function acts as the main indicator of bone healing progress. To improve prediction accuracy, a lead–lag compensator is designed and implemented to reduce the phase lag, effectively making the system respond faster. Incorporating control-theoretic principles into medical imaging is a novel aspect of this work, enabling the estimation of healing time without the need for serial radiographs. Real-time experiments on tibia fractures show that the estimated healing time closely matches clinician-recorded recovery periods. The method proves accurate across different fracture types when predicting healing times. Besides its precision, this new technique offers benefits such as reduced radiation exposure, cost savings, and clinical practicality.
BibTeX
@inproceedings{sridevi2025intelligent,
  author = {M. Sridevi and Svetha Venkatesh and P. Madhavasarma and P. Veeraragavan and Sylvain Ravel},
  title = {{An Intelligent Hybrid System for Fracture Detection and Healing Time Estimation from X- ray Images Using Lead-Lag CompensatorAnalyser}},
  booktitle = {Research Square},
  year = {2025},
  doi = {10.21203/rs.3.rs-8037369/v1},
}
Figure