Forward genetic screening links genotype to phenotype by introducing random genetic perturbations and identifying phenotype-altering mutations. Although genome-scale screens are routine for simple phenotypes in cultured cells, extending them to complex image-based phenotypes remains challenging. Here, we present spatially resolved CRISPR screening (SPARCS), a microscopy-based platform for forward genetic screening on single-cell images. SPARCS physically isolates mutants in situ by automated laser microdissection, enabling image-based screening at unprecedented scale with multimodal hit phenotyping. We demonstrate SPARCS in genome-wide CRISPR knockout screens of autophagosome formation and activation of the immune sensor STING across 70 million cells. Via machine learning-based image analysis, SPARCS recovered most macroautophagy genes and identified GPHR as a pH-dependent regulator of STING. Mass spectrometry-based proteomics of isolated hit cells revealed endoplasmic reticulum (ER)/Golgi disruption and nominated additional STING regulators via in silico perturbation modeling. These results establish SPARCS as a scalable platform for genome-wide genetic screening of complex cellular phenotypes with a proteome-level readout.