🪞 MirrorScan — AI-Powered Muscle Weakness Screening via Smart Mirror

The Problem

Your Solution

1. Scans your body posture and movement using computer vision (MediaPipe / YOLOv8 pose estimation)
2. Detects asymmetries, postural compensations, and movement quality linked to muscle weakness
3. Prompts 3 guided micro-movements (e.g., single-leg stand, squat, arm raise) lasting ~60 seconds
4. Generates a Muscle Weakness Index (MWI) — a simple score flagging early weakness patterns
5. Recommends action — exercises, physio referral, or GP visit if risk is high

Why It's Different (Your Differentiators)

How It Works (Tech Stack)

• Computer Vision: YOLOv8 pose estimation or MediaPipe BlazePose — detects 33 body keypoints in real time
• Movement Analysis: Measures joint angles, bilateral symmetry, movement speed, tremor, and range of motion
• ML Model: Trained on validated clinical data (SPPB, grip strength correlates) to map movement patterns → weakness grade
• Edge AI: Runs on-device (no cloud = privacy-safe), e.g., Raspberry Pi 5 or NVIDIA Jetson
• Output: Visual overlay on mirror + app report

Clinical Backing

• Validated movement tests (Timed-Up-and-Go, 5x Sit-to-Stand) can be translated to camera-based analysis — already proven in published research (PMC, Frontiers Neurology 2025)
• AI pose-based health monitoring is an active, growing research area with strong clinical validation

Target Users (Go-to-Market)

• B2C: Smart home users, fitness-conscious adults 40+
• B2B: Gyms, physiotherapy clinics, senior living facilities, corporate wellness programs
• B2B2C: Insurance companies offering preventive health incentives

Impact Pitch (1-liner for judges)

"We put a physio in every bathroom mirror — detecting muscle weakness before it becomes a disability."

Gaps to Address in Your Pitch

• Regulatory: Is this a wellness device (no FDA/CE needed) or a diagnostic tool (needs approval)? Position it as wellness screening initially.
• Data privacy: On-device AI = no video leaves the mirror. Lead with this.
• Validation plan: Partner with a physio clinic for a 20-person pilot study.

🛠️ How to Develop MirrorScan — Complete Guide

PHASE 1: Ideathon Prototype (48–72 hrs)

What to build

Step-by-Step Build Plan

Step 1 — Setup (2 hrs)

pip install mediapipe opencv-python numpy

• Use MediaPipe BlazePose (free, runs locally, detects 33 body landmarks in real time)
• Flip the webcam feed horizontally → instant "mirror" effect

Step 2 — Pose Detection (3 hrs)

import cv2, mediapipe as mp

mp_pose = mp.solutions.pose
cap = cv2.VideoCapture(0)

with mp_pose.Pose() as pose:
    while cap.isOpened():
        ret, frame = cap.read()
        frame = cv2.flip(frame, 1)  # Mirror flip
        results = pose.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
        # Draw landmarks on frame

Step 3 — Weakness Indicators (4 hrs)

Step 4 — Muscle Weakness Index Score (3 hrs)

def calculate_MWI(knee_angle, symmetry_diff, tilt_angle, speed):
    score = 100
    if knee_angle < 90: score -= 20      # Can't squat deep enough
    if symmetry_diff > 5: score -= 15    # Visible asymmetry
    if tilt_angle > 8: score -= 15       # Spine tilt = core weakness
    if speed > 3.0: score -= 20          # Slow sit-to-stand
    return score  # 0-100, higher = stronger

Step 5 — Visual Output (2 hrs)

• Color-coded skeleton (green = good, red = weak zone)
• Score card: MWI: 72 / 100 — Moderate Risk
• Simple text recommendation: "Left knee alignment weak — consider physio assessment"

PHASE 2: Tech Roadmap (Post-Ideathon)

Month 1-2:  Data collection → film 50+ people doing 5 standard movements
Month 3-4:  Train ML model (Random Forest / LSTM on keypoint time-series)
Month 5-6:  Validate against clinical gold standard (grip strength, SPPB scores)
Month 7-9:  Hardware — embed Raspberry Pi 5 + camera into mirror frame
Month 10-12: Pilot with 1 physio clinic or gym (20 users)
Year 2:     CE/FDA wellness device clearance + scale

ML Model (when you go beyond rule-based)

• Input: Time-series of 33 keypoints × (x, y, confidence) over 5 seconds
• Model: LSTM or Transformer for temporal movement patterns
• Labels: Matched against validated clinical weakness scores
• Training data: NTU RGB+D dataset (public), or your own collected data

PHASE 3: Pitch Deck Structure (for Judges)

Team Role Split (Mixed Multidisciplinary)

Free Tools & Resources

The Ideathon Demo Flow (2 minutes)

1. Person stands in front of webcam
2. Screen prompts: "Do 5 squats slowly"
3. Skeleton overlay appears in real time
4. Score appears: MWI: 68 / 100 — Left quad asymmetry detected
5. Recommendation: "Consult a physiotherapist for left knee assessment"

🪞 MirrorScan — Real-Time Muscle Weakness Highlighter

The Core Experience

You stand in front of the mirror → bend your knee → the quadriceps region glows RED if weakness is detected, BLUE if normal → you see it live on your own body, like an X-ray heat map.

How the Color Overlay Works

You (live camera feed)
        +
Pose skeleton (MediaPipe — 33 keypoints)
        +
Muscle zone mapping (keypoints → muscle region polygons)
        +
Weakness detection logic (angle, speed, symmetry)
        =
Colored muscle highlight drawn ON your body in real time

Step-by-Step Technical Flow

1. Detect Your Body Keypoints

2. Map Keypoints → Muscle Zones

3. Color Logic

Knee angle < 90° AND movement slow → RED on quad region (weakness)
Knee angle good, symmetric       → BLUE on quad region (normal)
One side weaker than other       → RED on that side only

4. Live Overlay on Mirror Feed

What It Looks Like (Visual)

┌─────────────────────────────┐
│                             │
│      👤 (you, live)         │
│                             │
│    [BLUE glow on left quad] │
│    [RED glow on right quad] │  ← right side weaker
│                             │
│  ⚠ Right Quadriceps Weak   │
│  Recommended: Physio check  │
└─────────────────────────────┘

Why This is Genuinely Innovative

• No wearables, no sensors, no clinic
• Muscle-level spatial feedback — not just a number score
• Feels like augmented reality on your own body
• Immediate, intuitive — even a non-medical person understands red = problem
• No product like this exists for general population screening today

For Your Ideathon Demo

1. Person stands in front of webcam (simulated mirror)
2. Does knee flexion slowly
3. Quadriceps zone lights up RED on weak side, BLUE on strong side
4. Text appears: "Right quadriceps weakness detected — Grade 2/5"
5. Recommendation shown below

Want me to write the actual Python code for this?

• ✅ Live webcam mirror feed
• ✅ MediaPipe pose detection
• ✅ Colored muscle zone overlays (red/blue polygons on body)
• ✅ Knee flexion weakness detection as the demo movement
• ✅ On-screen diagnosis text

Does It Already Exist? — Short Answer: Partially, but NOT exactly what you described.

✅ What EXISTS already

❌ What does NOT exist

A consumer-facing smart mirror that:

• Needs zero sensors or wearables
• Does real-time passive screening (just stand there and move)
• Highlights specific muscle zones on your actual body in red/blue
• Works for the general public (not just rehab patients)
• Gives a weakness grade from a simple natural movement

🎯 Your Innovation Gap (What Makes This Novel)

"The individual technologies exist. Nobody has put them together in a bathroom mirror for everyday people."

✅ Is It Possible? YES. Here's Why & How.

The Core Science Behind It

How the System Works — Step by Step

STEP 1: You stand in front of the mirror
         ↓
STEP 2: Camera captures your live video feed
         ↓
STEP 3: AI detects 33 points on your body
        (joints: shoulder, elbow, hip, knee, ankle...)
         ↓
STEP 4: System prompts a simple movement
        "Please bend your knee slowly"
         ↓
STEP 5: AI measures:
        • Knee angle (normal = 90°+)
        • Speed of movement
        • Left vs right symmetry
        • Spine alignment during movement
         ↓
STEP 6: Weakness detected → muscle zone glows RED
        Normal → muscle zone glows BLUE
         ↓
STEP 7: Mirror shows result on your body
        "Right quadriceps — Grade 3 weakness"

What the Camera Actually "Sees"

Normal knee flexion:
  → Smooth, reaches 90°, both sides equal → BLUE

Weak knee flexion:
  → Jerky, only reaches 60°, right side slower → RED on right quad

The Color Overlay — How It Appears on Your Body

Quadriceps = polygon drawn between:
  Hip point → Knee point → (width of thigh)

If weak → fills that polygon with RED, 40% transparent
If normal → fills with BLUE, 40% transparent

Is the Detection Accurate Enough?

What It CANNOT Do (Be Honest About This)

One-Line Technical Proof

MediaPipe runs at 30 frames/second, detects joints with sub-5° accuracy, and runs on a Raspberry Pi — all the hardware you need is already cheap and available.

Bottom Line

Idea Scorecard for Ideathon Judges

What Makes It Stand Out Specifically

The 2 Weak Points to Be Ready For

"We use the same joint-angle measurements physiotherapists use manually. Published research shows camera-based goniometry is within 5° of clinical tools — sufficient for screening, not diagnosis."

"It's positioned as a screening flag, like a home blood pressure cuff. It tells you 'consult a physio' — it doesn't tell you what's wrong. The liability sits the same way."

Compared to Typical Ideathon Submissions

• Another mental health chatbot
• A medication reminder app
• A telemedicine platform

• A hardware-software fusion with live AR demo
• A novel clinical use case with no direct competitor
• Something judges can physically interact with

Can It Detect Weakness Accurately? — Real Talk

The Honest Answer: It detects weakness indicators, not weakness itself.

What the camera sees:        What it actually means:
─────────────────────────────────────────────────
Knee angle = 55° on squat  →  Possible quad weakness
                               OR tight hip flexors
                               OR knee pain
                               OR just bad technique

How Accurate Is That Proxy?

The 3 Accuracy Challenges (Be Honest With Judges)

How Real Products Handle This

What This Means for Your Idea

The Blood Pressure Cuff Analogy

• Catches worrying trends early
• Motivates people to see a doctor
• Has saved millions of lives

Bottom Line on Accuracy

MMT Grades vs. What a Camera Can See

The Standard MMT Scale (MRC / Daniels & Worthington)

What Your Mirror CAN vs. CANNOT Detect

So Your Mirror Realistically Works for Grades 2–3

How to Map Grade 2 vs Grade 3 on Camera:

MMT Grade 3 (Fair) — Camera detects:
  → Person stands, tries to straighten knee from 90° bend
  → Achieves FULL extension (0°) against gravity
  → But slow speed (>3 seconds) or slight trunk lean = flag
  → Color: YELLOW (borderline)

MMT Grade 2 (Poor) — Camera detects:
  → Person cannot fully extend knee while standing
  → Only achieves partial range (stops at 30–40°)
  → Or knee wobbles/caves inward during attempt
  → Color: RED (weak)

MMT Grade 5 (Normal) — Camera baseline:
  → Full extension, fast, symmetric both sides
  → Color: BLUE (normal)

The 5 Camera-Detectable Weakness Signals (Mapped to MMT)

Your Mirror's MMT Detection Strategy

Honest MMT Limitation for Your Pitch

Grade 4 and 5 cannot be distinguished by camera alone — because both require resistance testing by a human hand. Your mirror covers Grade 0–3 detection, which represents the most clinically urgent cases needing early intervention.

"We detect MMT Grades 2–3 — the silent zone where people feel 'a bit weak' but haven't yet fallen or been diagnosed. That's 80% of the at-risk population we're targeting."

🪞 Where Else Can This Mirror Technology Be Used?

🏥 Clinical / Healthcare Applications

🧠 Neurological Screening

🏋️ Sports & Fitness

👶 Paediatric Screening

👴 Elderly Care / Geriatrics

💊 Chronic Disease Monitoring

🏫 Community / Public Health

Which One Should YOU Choose for Ideathon?

My Top 3 Recommendations

🥇 1. Parkinson's Early Detection Mirror

Most people are diagnosed 5–10 years after the disease starts. Camera detects tremor, facial expression reduction, arm swing asymmetry — all visible, all pre-diagnostic signs. Huge impact, no competitor.

🥈 2. Scoliosis Screening for School Children

Currently done manually once a year by a school nurse. Your mirror does it daily, automatically, with zero radiation (vs X-ray). Mass screening at scale. Parents love it. Schools love it.

🥉 3. Fall Risk Daily Check (Elderly)

Every morning the elderly person stands in front of bathroom mirror for 30 seconds → system silently checks balance → alerts family/doctor if risk increases. Massive market, emotional pitch.

🧬 Gene + Physiotherapy — Ideathon Ideas

🔬 What's the Link Between Genes & Physio?

• How fast your muscles recover after injury
• Your injury risk (ligament laxity, tendon stiffness)
• How you respond to exercise (strength vs endurance)
• Pain sensitivity (some people feel more pain genetically)
• Inflammation levels after exercise

💡 Top Ideas

🥇 Idea 1: GenoRehab — DNA-Guided Physiotherapy Prescription

• Correct exercise type (strength vs flexibility vs endurance)
• Recovery time between sessions
• Injury risk flags
• Pain management approach

🥈 Idea 2: GeneGuard — Genetic Injury Risk Screening for Athletes

🥉 Idea 3: NeurogenRehab — Gene-Based Stroke Recovery Optimization

• BDNF Met/Met carriers → need MORE repetitions, MORE time, different motor learning approach
• BDNF Val/Val carriers → respond faster to standard rehab

💡 Idea 4: PainGenix — Genetic Pain Profiling for Physio

• High pain sensitivity genotype → gentler progression, more pain management
• Low pain sensitivity genotype → can push harder, faster recovery

💡 Idea 5: Mirror + Gene = GeneMotion (Combines Your Original Idea!)

• Mirror detects: "Right quad weakness detected"
• Gene data adds: "COL5A1 variant — hyperlax ligaments; ACTN3 RR — power fiber type"
• System says: "Weakness likely structural + genetic. Recommended: eccentric strengthening, avoid passive stretching"

Comparison Table

My Recommendation

• It builds on your existing idea (not starting from scratch)
• It solves the accuracy problem of the mirror alone (gene data explains why)
• It's a two-layer innovation — nobody else will have this combination
• The pitch is powerful: "We don't just show you what's weak — we tell you why it's weak and exactly how to fix it, based on your DNA"

🤖 AI Ideas That Are Truly Critical for Physiotherapy

First — What Are the REAL Problems in Physio Today?

💡 The Ideas

🥇 1. AI Home Exercise Supervisor

70% of patients do their home exercises incorrectly. Nobody is watching. The physio finds out 2 weeks later when progress is poor.

• Patient opens phone camera at home
• AI watches them do their prescribed exercises
• Real-time correction: "Bend your knee more", "Keep your back straight"
• Automatically logs: reps done ✅, form quality score, pain level
• Physio gets a daily report without calling the patient

🥈 2. AI Re-injury Prediction & Early Warning System

Patients feel better → overdo it → re-injure → back to square one. This cycle costs months of recovery.

• Tracks movement quality daily via phone camera
• Builds a baseline of the patient's movement pattern
• Detects subtle deviations from baseline (compensations, reduced ROM, asymmetries)
• Sends alert: "Your gait pattern changed 15% in last 3 days — risk of re-injury. Rest today, contact your physio."

🥉 3. AI Clinical Decision Support for Physio Diagnosis

A new patient walks in with shoulder pain. The physio has 30 minutes. They run some manual tests. But differential diagnosis of shoulder pain has 12+ possible causes — rotator cuff, labrum, AC joint, referred cervical, impingement, frozen shoulder...

• Physio inputs: symptoms, ROM measurements, special test results (Hawkins, Neer, empty can, etc.)
• AI cross-references clinical decision rules + published evidence
• Generates: ranked differential diagnosis + recommended treatment pathway + red flags
• Like a clinical co-pilot — not replacing the physio, but making sure nothing is missed

💡 4. AI Rehab Outcome Predictor

Every patient asks: "When will I be better?" Physios guess based on experience. There's no objective predictor.

• Inputs: age, diagnosis, baseline ROM, pain score, functional tests, adherence history
• AI predicts: expected recovery timeline, probability of full vs partial recovery, risk of chronic pain
• Updates prediction weekly as patient progresses
• Helps physios flag patients who are recovering slower than predicted (may need imaging, specialist referral)

💡 5. AI-Powered Triage for Musculoskeletal Complaints

In most countries, the waiting list for physio is 6–12 weeks. By the time patients are seen, acute injuries have become chronic.

• Patient describes symptoms via app (voice or text)
• AI triages: Urgent (needs physio within 48hrs) vs Routine (can wait) vs Red Flag (needs immediate doctor referral)
• Provides immediate self-management advice for non-urgent cases
• Prioritizes booking queue automatically

Comparison: Importance vs Novelty

My Top Pick for Your Ideathon

🏆 AI Re-injury Early Warning System

• ❗ Problem is universally relatable — everyone knows someone who re-injured during recovery
• 🔬 Scientifically grounded — movement pattern changes before re-injury is clinically proven
• 📱 Buildable with just a phone camera — no hardware needed
• 💥 Nobody is doing this specifically — existing tools track progress, none predict re-injury
• 🎤 Pitch is emotional: "We catch the injury before it happens — not after"

🎯 Facial Palsy + Similar AI Scan Ideas — The Full Landscape

First — Your Facial Palsy Idea: Is It Novel?

• Only in hospital/clinical settings with expensive setups
• Requires trained clinicians to operate
• No home-use, real-time, patient-facing product exists
• No physiotherapy-integrated rehab tracking tool exists

🧠 How Facial Palsy AI Scan Works

AI detects 68–478 facial landmark points
         ↓
Measures LEFT vs RIGHT asymmetry of:
  • Eye closure (orbicularis oculi)
  • Brow raise (frontalis)
  • Smile symmetry (zygomaticus major)
  • Lip pursing (orbicularis oris)
  • Nasolabial fold depth
         ↓
Maps to House-Brackmann Grade (I–VI)
         ↓
Highlights weak muscle zones in RED on face

🔴 The Clinical Grading System (House-Brackmann Scale)

💡 Similar Ideas — Same Concept, Different Conditions

🥇 1. Bell's Palsy / Facial Palsy Rehab Tracker

• Patient scans face each morning
• App shows: "Orbicularis oculi (eye muscle) — Grade IV weakness" highlighted in red
• Prescribes targeted exercises: eye closure resistance, brow raises, lip stretches
• Tracks recovery weekly with objective improvement graphs
• Physio gets weekly progress report remotely

🔑 Clinically perfect — facial muscles are superficial, fully visible, and directly map to function. Best AI-detectable condition in all of physio.

🥈 2. Stroke Face + Limb Combo Scanner

• Combines facial asymmetry detection WITH arm drift test (F.A.S.T. test)
• Post-stroke rehab: tracks both facial and upper limb recovery together
• Can detect early stroke signs in high-risk patients (AF, hypertension)
• One scan covers two rehab domains

🥉 3. Cerebral Palsy Movement Scanner (Children)

• CP causes specific muscle overactivation (spasticity) — visible as abnormal posture/gait
• Camera detects: scissor gait, toe walking, arm posturing
• Highlights spastic muscle groups in red
• Guides targeted stretching + Botox decision support for clinicians

💡 4. Temporomandibular Joint (TMJ) / Jaw Muscle Scanner

• Patient opens/closes mouth, moves jaw side to side
• AI detects: deviation on opening (pterygoid weakness), limited range, asymmetry
• Common in dental patients, trauma patients, post-radiotherapy head & neck cancer
• No product exists for this at all

💡 5. Swallowing / Dysphagia Detection (Throat Muscles)

• Visible on camera: laryngeal elevation during swallow, neck muscle movement
• Relevant for: stroke, Parkinson's, head & neck cancer, elderly
• Current gold standard = videofluoroscopy (X-ray swallow) — expensive, radiation
• AI camera alternative = massive clinical need

💡 6. Eye Movement Disorder Scanner (Cranial Nerve Palsy)

• Patient follows a dot on screen
• AI tracks: limited abduction (CN VI palsy), ptosis (CN III), nystagmus
• Relevant for: head trauma, brain tumours, MS, diabetic neuropathy
• Currently assessed manually in clinic — no home monitoring exists

💡 7. Shoulder Muscle Wasting Scanner (Post-Nerve Injury)

• Muscle wasting (atrophy) changes the visible shape of the shoulder
• Camera detects: hollowing above/below spine of scapula = rotator cuff wasting
• Relevant for: brachial plexus injuries, rotator cuff tears, thoracic outlet syndrome

🏆 Ranking for Ideathon

My #1 Pick for Your Ideathon

🏆 Facial Palsy Rehab Tracker — "FaceRehab AI"

1. ✅ Face is perfectly visible to a camera — no clothing, no depth issues
2. ✅ 95% AI accuracy already proven in research
3. ✅ House-Brackmann grading = clear clinical scale to map to
4. ✅ 30,000+ new Bell's palsy cases every year in India alone
5. ✅ Current treatment = physio tells patient exercises → patient goes home → does them wrong → no tracking
6. ✅ Demo is instant — point camera at face, highlights appear on muscles immediately
7. ✅ Emotional pitch — Bell's palsy affects your smile, your face — deeply personal

"Bell's palsy takes your smile. FaceRehab AI gives it back — by tracking every facial muscle's recovery in real time, at home, every day."

📷 AI Scan-Based Clinical Assessment Ideas — Full Framework

The Core Concept You're Building Toward:

"Point a camera at a patient → AI automatically assigns a validated clinical score → no manual testing needed"

🧠 Brunnstrom Staging — How AI Can Assess It

What Brunnstrom Stages Are:

How AI Assigns the Stage:

Patient does 4 standard movements on camera:
  1. Raise arm forward (shoulder flexion)
  2. Touch opposite shoulder
  3. Hand behind back
  4. Finger-to-nose test
         ↓
AI measures:
  • Which joints move independently vs. together (synergy)
  • Speed, smoothness, range
  • Tremor / spasticity signs
         ↓
Output: "Brunnstrom Stage 3 — Upper Limb"
  + Color map: which movements are affected
  + Progress graph since last scan

💡 Other Clinical Scales That Can Be AI-Scanned — Same Concept

🥇 1. Fugl-Meyer Assessment (FMA) — Stroke Upper & Lower Limb

• Gold standard stroke motor recovery scale
• 33 upper limb items, 17 lower limb items
• All based on observable movement quality
• Research already shows depth camera can automate FMA scoring (published 2016–2025)
• Your idea: Automate it on a simple phone camera, no depth sensor needed

🥈 2. Modified Ashworth Scale (MAS) — Spasticity

• Measures muscle spasticity (resistance to passive movement)
• Currently: physio moves patient's limb and feels resistance
• AI version: Camera tracks limb velocity during movement + detects sudden "catch" (clasp-knife response) → assigns MAS grade
• Gap: Nobody has built a camera-only version yet

🥉 3. Functional Independence Measure (FIM) — ADL Assessment

• Measures how independently a patient can do daily activities
• AI version: Patient does 7 tasks on camera (transfers, grooming, eating simulation)
• AI scores each task 1–7 → generates full FIM score automatically
• Currently takes a physio 45 minutes to assess manually

💡 4. Gait Analysis Scales (FAC / TUGT / 10MWT)

💡 5. Pediatric Motor Scales — GMFCS (Cerebral Palsy)

• Gross Motor Function Classification System — 5 levels of CP motor function
• Currently: therapist observes child for 30+ minutes
• AI scan: Child plays for 5 minutes → AI classifies GMFCS level from movement patterns
• Huge need in low-resource countries where CP specialists are scarce

💡 6. ASIA Scale — Spinal Cord Injury Assessment

• American Spinal Injury Association impairment scale (A–E)
• Based on movement and sensation patterns
• AI version: Camera detects which body segments have voluntary movement → auto-assigns ASIA level
• Critical for SCI rehab planning

💡 7. Postural Assessment Scale for Stroke (PASS)

• 12 items measuring sitting and standing balance after stroke
• AI scan: Patient sits/stands on camera → AI scores each balance item automatically
• Currently takes 10–15 mins manual assessment

🏆 The Mega Idea — "RehabScan AI"

Patient stands/sits in front of camera → selects condition → does guided movements → AI automatically calculates the clinical score for that condition

Why This Is the Most Important Physio AI Idea

For Your Ideathon — Start with ONE Scale

• ✅ You clearly know it clinically
• ✅ All 6 stages are visually distinct
• ✅ Stroke rehab = massive global problem (13 million new strokes/year)
• ✅ No camera-only Brunnstrom auto-grader exists as a product
• ✅ Demo is clear: patient moves arm → app says "Stage 3" instantly