The study introduces ANORAK, an AI-based model that improves the histopathological grading of lung adenocarcinoma (LUAD). ANORAK, a pixel-wise segmentation method, encodes multiresolution inputs with an attention mechanism to delineate growth patterns from hematoxylin and eosin-stained slides. The model was evaluated in four independent cohorts, including 1,372 LUAD cases, and showed prognostic value for disease-free survival (DFS). AI-based grading consistently improved prognostication, particularly for stage I tumors, and assisted pathologists by improving the accuracy of prognostication in these tumors. Tumors with discrepant patterns between AI and pathologists had higher intratumoral heterogeneity. ANORAK also facilitated the morphological and spatial assessment of acinar patterns, capturing variations and transitions. The study highlights the precision quantification and morphology investigation of growth patterns, reflecting intratumoral histological transitions in LUAD.The study introduces ANORAK, an AI-based model that improves the histopathological grading of lung adenocarcinoma (LUAD). ANORAK, a pixel-wise segmentation method, encodes multiresolution inputs with an attention mechanism to delineate growth patterns from hematoxylin and eosin-stained slides. The model was evaluated in four independent cohorts, including 1,372 LUAD cases, and showed prognostic value for disease-free survival (DFS). AI-based grading consistently improved prognostication, particularly for stage I tumors, and assisted pathologists by improving the accuracy of prognostication in these tumors. Tumors with discrepant patterns between AI and pathologists had higher intratumoral heterogeneity. ANORAK also facilitated the morphological and spatial assessment of acinar patterns, capturing variations and transitions. The study highlights the precision quantification and morphology investigation of growth patterns, reflecting intratumoral histological transitions in LUAD.