KTS - Challenge #2 • The Body Scan - ElfieCare

## Inspiration

At ElfieCare, we believe preventive care should be simple, accessible, and actionable. Yet for most patients, tracking body composition still requires expensive devices, in-person visits, or complicated measurements that are hard to repeat consistently at home.

Challenge #2, The Body Scan, immediately resonated with us because it addresses a real and underserved need: giving adults a simple, free, and privacy-conscious way to estimate body measurements and body fat ratio using only two images, one front view and one profile view. We were inspired to build a solution that transforms everyday smartphone photos into meaningful health insights and helps patients visualize progress over time.

## What it does

KTS is an AI-powered body scan system that ingests two adult body images, one front and one side/profile, and produces:

• Estimated body measurements such as waist, hips, chest, and other key body dimensions
• Estimated body fat ratio
• A visual body-shape representation that helps users track improvement over time
• A structured result that can be used inside preventive care and digital health workflows

The product is designed to be easy to use: the user uploads two guided photos, and the system returns an instant assessment that is understandable for patients while still being useful for health coaching and monitoring.

## How we built it

We built KTS as a computer vision pipeline optimized for practical body-shape estimation from limited image input.

Our system combines:

• Image quality checks to validate whether the front and profile photos are usable
• Human body detection, pose estimation, and body segmentation to isolate the subject from the background
• Geometric feature extraction from body contours and key landmarks
• Anthropometric modeling to estimate circumference and linear body measurements from 2D views
• A machine learning regression layer to estimate body fat ratio from extracted body-shape features
• A visualization layer to show body-shape trends and progress over time

To improve real-world usability, we designed the workflow to handle common consumer-grade inputs such as smartphone photos taken at home. We also focused on making the system modular, so each component, quality control, landmarking, measurement estimation, and body fat prediction, can be improved independently as more data becomes available.

## Challenges we ran into

One of the biggest challenges was the inherent ambiguity of estimating 3D body characteristics from only two 2D images. Small differences in camera angle, lighting, posture, distance, and clothing can significantly affect body contours and downstream predictions.

Another major challenge was balancing accessibility with reliability. We wanted users to take photos easily at home, but accurate estimation requires some consistency in stance and framing. This forced us to think carefully about input guidance, quality validation, and uncertainty handling.

We also had to consider privacy and trust. Body images are highly sensitive, so we approached the solution with the assumption that secure processing and minimal-friction user consent are essential for adoption in healthcare settings.

## Accomplishments that we're proud of

We are proud that KTS turns a difficult technical problem into a simple user experience: two images in, useful health insights out.

We are especially proud of:

• Building an end-to-end prototype that addresses both required dimensions of the challenge: body measurement estimation and body fat ratio estimation
• Designing the system around real patient accessibility instead of specialized hardware
• Creating a foundation for longitudinal progress tracking, not just one-time prediction
• Aligning the product with ElfieCare’s preventive care mission, where low- cost, repeatable body monitoring can support healthier behavior over time

## What we learned

This challenge reinforced that health AI is not only about model performance; it is also about usability, trust, and consistency of input.

We learned that:

• Input quality control is just as important as the prediction model itself
• Body composition estimation benefits from combining geometric reasoning with machine learning rather than relying on a single black-box model
• Users need clear guidance and understandable outputs if they are expected to use the product regularly
• In healthcare, a practical and repeatable solution can create more value than a technically complex solution that users cannot adopt easily

## What's next for KTS - Challenge #2 • The Body Scan - ElfieCare

Our next steps are focused on improving accuracy, robustness, and clinical usefulness.

We plan to:

• Expand validation on more diverse body types, lighting conditions, and clothing scenarios
• Improve calibration and uncertainty estimation so users know when a scan is reliable
• Add stronger visual progress tracking to help patients see body-shape changes over time
• Integrate the system more deeply into ElfieCare’s care journeys for prevention, coaching, and follow-up
• Explore privacy-preserving deployment options for safer real-world adoption

Our long-term vision is to make body composition tracking as easy as taking two photos, empowering patients with simple, repeatable, and affordable insights that support better health decisions.

Built With

• c++
• dart
• flutter
• nextjs
• ruby