Pixaire1: Evaluation of automated chronic wound surface measurement systems.

This study evaluates two smartphone-based wound measurement systems, finding that the semi-automated Woundtrack method offers accuracy and precision comparable to the reference digitized planimetry and is highly effective for clinical use, while the fully automated Woundsize method, though generally reliable, shows greater discrepancies in small wounds and is best implemented with clinician validation.

The Gist

The Big Picture: Measuring the "Painful Puzzle"

Imagine you have a chronic wound (like a stubborn sore that won't heal). To know if it's getting better, doctors need to measure its size. It's like trying to track the growth of a garden patch: if you don't measure it accurately, you might think it's shrinking when it's actually spreading, or vice versa.

For a long time, the "gold standard" for measuring these wounds was Digitized Planimetry (PL). Think of this as the "old-school, high-precision method." A doctor places a clear plastic sheet over the wound, traces the outline with a pen, takes a photo of that tracing, and uses computer software to calculate the exact area. It's accurate, but it's slow, messy, and requires special tools.

The Goal of this Study:
The researchers wanted to see if two new smartphone apps could do the same job, but faster and easier. They tested two "digital assistants":

1. Woundtrack (WT): The "Semi-Automatic Assistant." The doctor looks at the wound on their phone screen and traces the outline with their finger. The app then does the math.
2. Woundsize (WS): The "Fully Automatic Assistant." The doctor just takes a picture, and an AI (Artificial Intelligence) tries to guess where the wound ends and healthy skin begins, then calculates the size automatically.

They compared these two apps against the "Gold Standard" (the plastic sheet tracing) to see who was the best.

---

The Experiment: A Race Between Three Methods

The study took place in a hospital in France with 42 patients. Two expert doctors measured the same wounds using all three methods (Plastic Sheet, Woundtrack, and Woundsize).

The Results in Plain English:

1. The "Semi-Automatic" Assistant (Woundtrack) 🏆

Verdict: The Reliable Partner.
This method was a huge success. When the doctor traced the wound on the phone, the result was almost identical to the old-school plastic sheet method.

• The Analogy: Imagine you are drawing a map. If you trace the border carefully, your map is perfect. Woundtrack is like a very helpful compass that ensures your map is accurate as long as you draw the line.
• Why it works: Because a human is still drawing the line, the AI doesn't get confused by weird lighting or strange skin colors. It just calculates the area of the shape you drew.
• The Catch: It works great on big wounds, but on tiny wounds (smaller than a credit card), even a tiny mistake in drawing the line can make the percentage error look huge. However, in absolute terms (real square centimeters), the difference was negligible.

2. The "Fully Automatic" Assistant (Woundsize) ⚠️

Verdict: The Talented but Flaky Intern.
This method was impressive but inconsistent. Sometimes it was perfect; other times, it got it wrong.

• The Analogy: Think of this AI as a very smart intern who has never seen a wound before. If the lighting is perfect, the intern does a great job. But if the room is too dark, too bright, or if there's a weird shadow, the intern gets confused and draws the wrong line.
• The Problem: The study found that the AI's biggest enemy wasn't the shape of the wound or the patient's skin color—it was bad lighting. If the photo was too dark (underexposed) or had a blinding flash (overexposed), the AI couldn't tell the difference between the wound and the healthy skin.
• The Solution: The researchers suggest we shouldn't let the AI work alone. Instead, use a "Propose and Correct" workflow: The AI makes a guess (proposes), and the doctor looks at it, fixes any mistakes, and hits "approve."

3. The "Gold Standard" (Plastic Sheet) 📏

Verdict: The Perfectionist.
As expected, the plastic sheet method was the most consistent. However, it's slow and requires physical contact with the wound, which isn't always practical for home care or quick check-ups.

---

Key Takeaways for Everyday Life

1. Smartphones are ready for prime time (mostly).
You don't need a $10,000 scanner to measure a wound anymore. A standard smartphone with the Woundtrack app is just as accurate as the old plastic sheet method. This means nurses and doctors can measure wounds quickly at the patient's bedside or even at home, saving time and reducing stress.

2. AI is a tool, not a replacement (yet).
The Woundsize (fully automatic) feature is cool, but it's not ready to drive the car alone. It needs a human driver (the doctor) to keep an eye on the road. If the photo is taken in bad light, the AI will fail. The best workflow is: AI guesses -> Human checks -> Human approves.

3. Small wounds are tricky for everyone.
Measuring very small wounds is hard for everyone, whether using a pen on plastic or a finger on a screen. A tiny error in drawing the line on a 1cm wound looks like a huge percentage error. But in reality, the actual difference in size is so small it doesn't really matter for treatment.

4. Lighting is everything.
If you want the automatic AI to work, you need good photos. No flash, no shadows, and good natural light. If the photo is dark, the AI gets lost.

The Bottom Line

The study concludes that Woundtrack is a fantastic, easy-to-use tool that is just as good as the traditional method. It's a win for patient care because it's faster and easier. Woundsize (the fully automatic one) shows promise but needs a human to double-check its work, especially if the photo isn't perfect.

In short: Use the phone to measure, but keep your eyes open to check the work!

Technical Summary

1. Problem Statement

Chronic wounds (e.g., diabetic foot ulcers, venous leg ulcers, pressure ulcers) require precise monitoring of surface area to track healing kinetics and adjust treatment. However, current clinical practice rarely utilizes quantitative surface measurement due to the time consumption and specialized equipment required by traditional methods.

• The Gap: There is a need for reliable, reproducible, and easy-to-access measurement tools that can be integrated into routine care (e.g., via smartphones) without sacrificing accuracy compared to the "gold standard."
• The Objective: To evaluate two smartphone-based algorithms developed by Pixacare:
  1. Woundtrack (WT): A semi-automated method where a clinician manually outlines the wound on a touchscreen.
  2. Woundsize (WS): A fully automated method using deep learning (Convolutional Neural Networks) to segment the wound.
  • Reference Standard: Digitized Planimetry (PL), involving manual tracing on acetate followed by digital analysis via ImageJ.

2. Methodology

Study Design:

• Type: Open-label, single-center, cross-sectional study.
• Location: Department of Vascular Surgery, CH de Haguenau, France.
• Timeline: May–June 2023.
• Population: 42 patients with chronic wounds (4–150 cm²).
  • Exclusions: 6 patients were excluded (4 due to multiplanar wounds, 2 due to indistinct edges), leaving 36 patients for final analysis.
• Data Collection: Two independent experts (Expert A and Expert B) measured each wound using three methods:
  1. PL (Reference): Transparent dressing applied, wound traced on acetate, digitized via ImageJ.
  2. WT (Semi-auto): Photo taken with iPhone 12 (with ArUco marker sticker for scale/color), clinician manually traces wound on the app.
  3. WS (Auto): Same photo as WT, processed by a ResNet-based deep learning algorithm for automatic segmentation.

Statistical Analysis:
A four-step analysis was performed:

1. Multivariate Analysis: Friedman's test and Nemenyi's post-hoc test to compare overall variance between the three methods.
2. Precision (Reproducibility): Intra-class correlation (ICC), Pearson correlation, and Bland-Altman plots to assess agreement between the two experts for each method.
3. Accuracy: Comparison of WT and WS against the PL reference using Student's t-test, linear regression, and ICC.
4. Non-Conformity Analysis: Specific analysis of wounds < 8 cm², defined as "non-conformity" if the deviation from PL exceeded 20%.

3. Key Contributions

• Validation of Semi-Automated Workflow: The study provides clinical validation for the Woundtrack algorithm, demonstrating it is a viable alternative to digitized planimetry for routine care.
• Identification of Automated Limitations: The study highlights that fully automated segmentation (Woundsize) is highly sensitive to image quality (specifically exposure/lighting) rather than algorithmic complexity.
• Proposed Hybrid Workflow: Instead of full automation, the authors propose a "Human-in-the-Loop" workflow: Algorithm proposes segmentation $\rightarrow$ Clinician corrects and validates.
• Supplementary Optimization: The authors demonstrated that using the Woundtrack algorithm on transparent acetate tracings (rather than direct photos) significantly improves accuracy for large/complex wounds by eliminating geometric projection errors.

4. Key Results

A. Overall Comparison (Multivariate Analysis)

• No statistically significant difference was found between the three methods overall (Friedman's test, $p = 0.82$).

B. Precision (Inter-rater Reliability)
All methods showed excellent reliability between the two experts:

• PL (Reference): ICC = 1.0 (Perfect).
• Woundtrack (WT): ICC = 0.96 (Excellent).
• Woundsize (WS): ICC = 0.94 (Excellent).
• Observation: WT showed slightly better consistency than WS, but both were highly reproducible.

C. Accuracy (vs. Reference Planimetry)

• Woundtrack: Showed excellent agreement with PL (ICC = 0.96, $p < 0.001$). Linear regression showed a slope of 1.3 with minimal bias.
• Woundsize: Also showed good agreement (ICC = 0.95), but with slightly higher variability than WT.

D. Small Wound Analysis (< 8 cm²)

• Woundtrack: 70.1% conformity. Statistical tests (Levene, Bartlett, Wilcoxon) showed no significant difference in variance or distribution between WT and PL errors.
• Woundsize: 63% conformity. Statistical tests showed a significant difference in variance between WS and PL ($p = 0.03$ for Levene's test), indicating WS is less reliable for small wounds.

E. Qualitative Analysis of Errors (Supplementary)

• Analysis of segmentation failures in Woundsize revealed that image exposure (over/under-exposure) was the primary predictor of failure (Odds Ratio = 19.15, $p < 0.001$).
• Wound complexity (maceration, hyperkeratosis) and foreign bodies were not statistically significant predictors of failure.
• Conclusion: The limitation lies in image acquisition quality, not the deep learning model itself.

5. Significance and Conclusion

Clinical Significance:

• Woundtrack is recommended as an effective, time-saving alternative to digitized planimetry. It offers a "real benefit" in implementation due to ease of use and speed, with accuracy comparable to the gold standard.
• Woundsize (fully automated) is not yet ready for autonomous clinical use, particularly for small wounds or in variable lighting conditions.

Recommendations:

1. Adopt Semi-Automation: Clinicians should use the semi-automated Woundtrack method for routine monitoring.
2. Hybrid Workflow for Automation: For automated tools, the most efficient implementation is a collaborative process where the AI proposes a segmentation, and the clinician validates/corrects it.
3. Image Quality Control: Strict protocols for image exposure are critical for any automated wound analysis system.
4. Handling Large/Complex Wounds: For wounds with complex 3D geometry, the study suggests a hybrid technique: trace the wound on a transparent dressing, then use the smartphone app to measure the tracing. This bypasses the geometric distortion inherent in photographing curved surfaces.

Limitations:

• The study lacked patients with high skin phototypes (Fitzpatrick 5-6), which could impact automated segmentation performance.
• Large wounds (>60 cm²) and multiplanar wounds present geometric challenges for 2D projection methods.