Home▸AI & Machine Learning▸Grad-CAM: Where a Medical-Scan AI Is Looking (2D)

Grad-CAM: Where a Medical-Scan AI Is Looking (2D)

Interactive 2D Class Activation Mapping simulator: a real, hand-specified convolutional network classifies a synthetic medical scan as Normal or Abnormal, and a genuine gradient-weighted class activation map (Grad-CAM) shows exactly which pixels drove that decision, laid out flat as input, heatmap and feature-map panels.

AI & Machine Learning2DAdvanced60 FPS📱 Mobile-adapted⇄ 3D version

Details

CategoryAI & Machine Learning

RenderingCanvas 2D

DifficultyAdvanced

Target FPS60

ControlsDrag sliders to move a synthetic lesion and adjust its contrast, toggle which class (Abnormal/Normal) the heatmap explains, and switch the CAM source between the final conv layer and the raw layer-1 blur filter — the flat 2D layout shows the input scan, CAM overlay, all layer-1 and layer-2 feature-map panels, and the output probability bars simultaneously, no orbiting required.

Grad-CAMexplainable AImedical imagingCNNinterpretability2D

Related simulations

SHAP Feature Attribution: Shapley Lattice & ConvergenceWatch the Shapley-value engine behind SHAP explanations in 2D: the entire Boolean coalition…

Inside AI-Assisted Diabetic Retinopathy ScreeningAn interactive 3D fundus camera and retina model that shows how a CNN scans a retinal…

How Deep Learning Reads a Chest X-Ray for Signs of PneumoniaAn interactive 3D walkthrough of a CNN reading a chest X-ray for pneumonia: watch the image… ⊞ All 1161 in AI & Machine Learning →

This 2D companion to the 3D Grad-CAM simulator runs the exact same kind of real, hand-specified two-layer convolutional network — actual 3×3 convolutions, ReLU, max-pooling, global average pooling and a linear head — over a synthetic scan containing a movable, adjustable-contrast lesion, then computes a genuine Class Activation Map: the gradient-weighted sum of the final convolutional layer's feature maps that reduces, for a global-average-pooled linear head, to the exact formula used in real Grad-CAM interpretability tooling. Along the way it fixes a real calibration bug found by numerically testing the 3D source: that sim's deep "lesion-selective" channel was wired to a blob/Laplacian filter that saturates on the background's high-frequency texture instead of the lesion, so its predicted class barely moved as contrast increased. Here the same channel is wired to a blur/local-brightness filter instead — verified to make the classifier's Normal/Abnormal call genuinely track the lesion — while the CAM math, panel layout and every other control stay faithful to the original two-layer architecture.

⚙ Under the hood

Interactive 2D companion to the 3D Grad-CAM simulator: a real two-layer convolutional network classifies a synthetic medical scan as Normal or Abnormal on flat heatmap panels, computing a genuine Class Activation Map, and fixes a calibration bug found in the 3D source where the lesion-selective channel saturated on background texture instead of tracking the lesion.

Grad-CAMexplainable AImedical imagingCNNinterpretability2D

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install