Medical Image Learning Game

Educational safety notice: Every image is generated inside your browser from abstract shapes and textures. This game does not diagnose, treat, or evaluate any medical condition.

Game setup

Game mode

Difficulty

Synthetic image theme

Random seed

Object count 5

Noise 18

Blur 1

Contrast 105%

Canvas workspace

Start the game to generate a synthetic learning mission.

Level 1 Classification

x: 0, y: 0 · synthetic data

0Score

0%Accuracy

0Streak

00:00Time

Answer confidence 70%

Choose a mode and press Start game.

Campaign progress 0 / 12

Image and annotation options

Brightness 100%

Rotation variation 20°

Occlusion 5%

Brush size 18

Class distribution

Round time

Accessibility and display

Artificial dataset builder

Collect generated examples, assign fictional labels, balance classes, and prepare training, validation, and test splits.

0 samples

Assigned label

Dataset split

Quality

Mark as difficult

Import JSON

ID	Label	Split	Quality	Difficulty	Seed	Actions
No artificial samples collected yet.

Simulated pattern model

Explore training behaviour without making medical predictions.

Idle

Epochs 20

Training samples 24

Learning rate 0.010

Batch size

Image resolution

Augmentation plan

Decision threshold 0.50

Model version

Individual augmentation options

Flips Rotation Crop and translation Synthetic noise Brightness and contrast Random erasing Pattern occlusion

—Validation accuracy

—Final loss

—Generalisation gap

—Simulated time

Add synthetic samples, then run the training simulation.

Learning dashboard

Review performance, confidence, timing, class balance, and artificial model behaviour.

0%Precision

0%Recall

0%F1 score

0%Specificity

0%Mean IoU

0%Confidence match

—Strongest fictional class

—Class needing practice

0.0sAverage completion time

0Personal best score

Complete a round to receive a recommended next lesson.

No achievements unlocked yet

Learning library

Pixels and resolution

Images are grids of values. Resolution changes visible detail and computational cost.

Brightness and contrast

Brightness shifts intensity. Contrast separates light and dark areas.

Shape and texture

Edges describe boundaries. Repeated local changes create texture clues.

Noise and blur

Noise adds random variation. Blur can hide small boundaries and details.

Classification

A classifier selects a label for an entire artificial image.

Localisation

Localisation finds where a target appears using a point or box.

Segmentation

Segmentation assigns a region label at pixel or brush-stroke level.

Data splits

Training fits patterns. Validation guides choices. Testing checks final behaviour.

Augmentation

Controlled transformations add variety. Excessive changes may destroy useful patterns.

Class imbalance

Uneven examples can bias results toward the most frequent class.

Confidence

Confidence should match correctness. High confidence errors deserve careful review.

Human review

Uncertain, unusual, or high-impact automated outputs should be reviewed by people.

Keyboard shortcuts and controls

S Start or resume

N Next sample

Enter Submit answer

B Bounding box tool

P Point tool

D Brush tool

H Pan tool

Ctrl + Z Undo

+ Zoom in

- Zoom out