Grad-CAM: Where a Medical-Scan AI Is Looking

Synthetic Lesion

Horizontal position 50%

Vertical position 50%

Lesion contrast 40%

Explain Class

CAM Source Layer

Classifier Output

P(Abnormal)

—

P(Normal)

—

Localization IoU vs true lesion

—

Peak raw CAM activation

—

How it works

A small hand-built CNN (fixed, non-random-init weights so results are reproducible) processes a synthetic 32×32 scan: two real convolutional layers (edge / blob / texture filters → ReLU → 2×2 max-pool), a global-average-pool, and a linear classifier head scoring "Normal" vs "Abnormal".

Class Activation Map (Zhou et al. 2016;
generalized as Grad-CAM by Selvaraju et al. 2017):

  L^c(i,j) = ReLU( Σ_k w_k^c · A_k(i,j) )

  A_k(i,j)  = k-th feature-map channel, final conv layer
  w_k^c     = ∂y^c / ∂(GAP(A_k))
            = classifier weight for channel k, class c
            (exact when the head is GAP + linear, as here)

Position / contrast sliders — move a real Gaussian-bump "lesion" through the image and control how strongly it stands out from the background texture.
Explain Class — chooses which class's score y^c is being explained; the weight vector w^c flips, so the same feature maps produce a different heatmap.
CAM Source Layer — Final conv shows the true gradient-weighted CAM (8×8, coarse but class-discriminative); Layer 1 shows one filter's raw response before any pooling (32×32, sharp but not class-aware) — the classic depth/resolution trade-off in CNN interpretability.
3D layout — input scan → CAM overlay (left), Layer 1 feature-map bars (middle), Layer 2 feature-map bars (right), output score bars (far right). Bar height and color both encode activation strength.

Real-world relevance: this is the exact technique radiologists and regulators demand from black-box scan classifiers (chest X-ray, mammography, dermoscopy AI) before trusting a "positive" call — the heatmap must land on the actual lesion, not on an unrelated artifact.

Low activation Mid activation High activation

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.