Artificial Intelligence Applications in Bone Scintigraphy: Where We Are Today

Article Title: From Image to Index: Diagnostic Accuracy of a Novel Semi-Quantitative Approach for Assessing Suspected Periprosthetic Joint Infection with Triple-Bone Phase Scintigraphy

Background

Total hip and knee arthroplasty are frequently performed procedures for patients with end-stage osteoarthritis who fail conservative therapy. However, these surgeries can result in serious complications, particularly aseptic loosening and periprosthetic joint infection (PJI), which is one of the most common causes of persistent postoperative pain. Differentiating between infection and aseptic loosening remains a diagnostic challenge, especially in cases with indeterminate clinical or imaging findings.

Triple-phase bone scintigraphy (3PBS) is a widely available nuclear medicine technique valued for its high sensitivity and its resistance to metallic artifacts that affect other modalities such as CT and MRI. Despite its strengths, 3PBS interpretation often depends on qualitative assessment, which is susceptible to interobserver variability. To address this limitation, the authors proposed a novel semi-quantitative metric, the Blood Pool-to-Delayed Ratio (BPrDr) variation index, aiming to provide a more objective and reproducible tool for assessing suspected periprosthetic infections.

What Does Previous Literature Tell Us?

• Bone Scintigraphy Based on Deep Learning Model and Modified Growth Optimizer
  A 2024 Scientific Reports study developed a machine learning technique combining Mobile Vision Transformer (MobileViT) with a Growth Optimizer enhanced by the Arithmetic Optimization Algorithm to analyze bone scintigraphy images. Using 2,800 scans, the algorithm outperformed other models, capturing both local textures and global patterns.

• AI-Based Analysis of Whole-Body Bone Scintigraphy
  A 2024 multicenter study in Zeitschrift für Medizinische Physik tested ten CNN models against three nuclear medicine physicians. For classifying normal vs. abnormal scans, DenseNet121 with attention aggregation performed best (AUC 0.72, accuracy 72%). For distinguishing malignant vs. non-neoplastic disease, InceptionResNetV2 with spatial pyramid pooling led (AUC 0.72, specificity 88%). AI matched or outperformed physicians while reducing interpretation time.

• Deep Neural Network–Based AI for Cancer Bone Metastasis
  A 2020 Scientific Reports study trained a DNN on 12,000 scans. Performance was excellent across cancers: AUC 0.988 (breast), 0.955 (prostate), 0.957 (lung). Against physicians on 400 cases, AI achieved higher accuracy (93.5% vs. 89%) and sensitivity (93.5% vs. 85%), with interpretation time ~11 seconds. Consulting the AI reduced false negatives and improved confidence.

Results

• 64 patients included: 17 with PJI, 47 with aseptic loosening.

• Mean BPrDr variation index higher in infected prostheses (33.4 ± 27.7) vs. aseptic (17.0 ± 14.2), p=0.031.

• ROC AUC: 0.71 (moderate discriminative ability).

• Optimal threshold: 14.73%.

  • Sensitivity: 88.2%

  • NPV: 92.8%

  • Specificity: 55.3%

  • PPV: 41.6%

Decision curve analysis showed net clinical benefit across thresholds.

Conclusion

The BPrDr variation index is a promising semi-quantitative tool to improve 3PBS diagnostic performance in suspected periprosthetic joint infections. While specificity was modest, its high sensitivity and negative predictive value suggest utility as a rule-out tool, helping avoid unnecessary invasive testing or advanced imaging in low-probability cases.

Editorial Comment

This study introduces a practical method to increase objectivity in 3PBS interpretation. The BPrDr index is easy to calculate using standard equipment and could reduce diagnostic uncertainty. However, retrospective design and small cohort size demand prospective validation.

Given moderate specificity, positive results should prompt confirmatory imaging (e.g., labeled leukocyte scintigraphy, FDG PET/CT). Still, the approach is attractive for screening in resource-limited settings where advanced modalities are unavailable.

Teaching Points for Residents

• Triple-phase bone scintigraphy is a valuable first-line modality for suspected PJI due to high sensitivity and artifact resistance.

• The BPrDr index quantifies hyperemia vs. delayed osteoblastic activity, aiding distinction of infection vs. loosening.

• At a 14.73% threshold, sensitivity (88%) and NPV (93%) support use as a rule-out test.

• Positive cases should be correlated with labs and clinical findings due to limited specificity.

• Adding semi-quantitative indices to reporting may enhance reproducibility and reduce unnecessary testing.