Longitudinal Wound Healing Tracker

• The tracker is a computational tool that quantifies wound healing progression by computing dynamic metrics like healing rate, total percent healing, and severity score.
• It integrates seamlessly with automated image segmentation pipelines, using AI models to extract wound areas and generate clinical alerts based on temporal trends.
• The system supports telemedicine and IoMT frameworks, enhancing chronic wound management by reducing subjectivity and enabling proactive intervention.

A longitudinal wound healing tracker is a computational component integrated into wound image analysis pipelines, enabling the quantification and temporal monitoring of wound healing progression. Its primary function is to compute dynamic, quantifiable measures of healing—such as healing rate, severity score, and alert generation—using time series of wound area measurements derived from clinical or telemedicine-acquired wound images. The longitudinal tracker is critical for real-time decision support in chronic wound management, underpins telemedicine frameworks, and supports reproducibility in AI–driven wound care (Kiprono, 20 Dec 2025).

1. Formal Definition and Core Objectives

A longitudinal wound healing tracker ingests serial wound area measurements $A_t$ at discrete time points $t$. It computes key progression metrics, such as instantaneous healing rate and total healing percentage, maps healing trajectories to standardized severity scores, and emits clinical alerts upon deviations from expected healing patterns. In the WoundNet-Ensemble system, these metrics are automatically computed following spatial segmentation and classification of wounds (Kiprono, 20 Dec 2025).

2. Mathematical Formulation of Healing Progression Metrics

The tracker computes several clinically interpretable indicators:

• Instantaneous Healing Rate: Defined as

$$\text{HealingRate}_t = \frac{A_{t-1} - A_t}{A_{t-1}} \times \frac{100}{\Delta t}$$

where $A_t$ is wound area at time $t$ (cm²), $\Delta t = t-t_{-1}$ (days). The output is percent per day.

• Total Percent Healing from Baseline:

$$\text{TotalHealing}_t = \frac{A_0 - A_t}{A_0} \times 100\%$$

where $A_0$ is the wound area at baseline.

• Severity Score: Discrete integer, empirically mapped (range 1–10) using area and healing rate; e.g., if $A_t > 25\,\text{cm}^2$ and healing rate $<2\%/$day, then severity score = 9.

Table: Healing Tracker Computed Metrics

Metric	Formula	Output
Healing Rate	$$\frac{A_{t-1} - A_t}{A_{t-1}} \times \frac{100}{\Delta t}$$	% per day
Total Percent Healing	$\frac{A_0 - A_t}{A_0} \times 100\%$	% since baseline
Severity Score	Integer 1–10, mapped from area and rate thresholds (empirical and protocol-defined)	Ordinal category

Clinical alerts are generated if any of the following are met:

• Healing rate $\leq 0$ (no reduction or increase in area)
• Severity score increases compared to prior timepoint
• Total healing below protocol threshold (e.g., $<10\%$ at day 7) (Kiprono, 20 Dec 2025).

3. Integration with Automated Image Analysis Pipelines

The longitudinal tracker operates downstream of wound segmentation and classification modules. In deployed AI pipelines such as WoundNet-Ensemble (Kiprono, 20 Dec 2025), segmentation models (e.g., dual-attention U-Net++ (Cieślak et al., 7 Jul 2025), LinkNet/U-Net ensembles (Mahbod et al., 2021)) predict $A_t$ by extracting the wound region from calibrated wound photographs. The tracker then updates healing metrics at each subsequent image capture. These metrics support temporal visualization, automated reporting, and clinical workflow integration.

In multi-site studies, wound area extraction accuracy is enhanced by rigorous segmentation architectures with Bayesian hyperparameter tuning and test time augmentation to ensure reliable time series (Cieślak et al., 7 Jul 2025, Mahbod et al., 2021). These methodological steps address variability due to lighting, pose, and device heterogeneity.

4. Deployment Modalities and IoMT Implementation

The tracker is a fundamental component in Internet of Medical Things (IoMT) systems designed for telemedicine and remote monitoring. In reference implementations (Kiprono, 20 Dec 2025):

• Images are acquired via mobile applications or edge devices, securely transmitted via encrypted channels (TLS, AES-256).
• Area extraction and healing tracking may occur on either local edge nodes (e.g., GPU/TPU-equipped gateways for near real-time feedback) or HIPAA-compliant cloud infrastructure.
• Resources: the WoundNet-Ensemble system achieves $\approx$46.4 ms per-image latency on RTX-class GPUs at $224 \times 224$ input size; longitudinal metric computation is computationally negligible compared to inference (Kiprono, 20 Dec 2025).

5. Clinical Relevance and Decision Support Functions

Objective wound healing quantification provides several advantages:

• Reduces subjectivity in wound assessment and mitigates inter-rater variability.
• Enables early detection of stagnation or deterioration (e.g., non-healing, exacerbation).
• Auto-generates notifications to prompt intervention, improving healing trajectories and potentially reducing adverse outcomes such as amputation or infection (Kiprono, 20 Dec 2025).
• Provides reproducible, time-stamped records for telehealth, EHR integration, and clinical trials.

A plausible implication is that robust trackers facilitate outcome audits, protocol compliance monitoring, and AI-driven triage in both hospital and community care.

6. Limitations and Future Research Directions

Documented limitations of current longitudinal trackers include:

• Dataset scope: Existing trackers may lack support for certain wound etiologies (e.g., arterial, surgical) due to training set constraints (Kiprono, 20 Dec 2025).
• Single-modality imaging: Relies on RGB photographic imagery, introducing bias across skin tones or wound environments.
• Absence of direct quantification of tissue composition, depth, or exudate, which could inform more sophisticated healing models (Anisuzzaman et al., 2022).

Future work includes:

• Multi-modal data integration (EHR data, physiological sensors, smart bandages) to augment trajectory modeling.
• Model compression/quantization to decrease inference/resource footprint, enabling truly offline or at-home use.
• Publishing implementation and trained weights to ensure reproducibility.
• Embedded randomized controlled trial protocols to validate clinical impact on healing time, amputation rates, and costs (Kiprono, 20 Dec 2025).

7. Comparative Context in the Literature

Earlier work emphasizes the inadequacy of single-image severity classification and subjective scoring, with lower accuracy, often omitting longitudinal analysis (Anisuzzaman et al., 2022). The gap addressed by algorithmic trackers is the robust, reproducible, and quantitative evaluation of healing trajectories—integrating classifier/segmentation pipelines with explicit temporal logic and protocol-derived alerting. Results on the largest benchmark (5,175 wound images, six types) show 99.90% accuracy for ensemble classification and demonstrate real-world deployability in clinical and telemedicine contexts (Kiprono, 20 Dec 2025). This performance substantially exceeds earlier methods, which focused on either single timepoint assessment or lower-performing single-model approaches (Anisuzzaman et al., 2022, Patel et al., 2023).