# Loss: supervised + cell continuity + sparsity loss = MSE(pred, y) + 0.01 * continuity_loss(model) + 0.001 * sparsity_penalty(cell_mask)