The artificial intelligence-based model ANORAK improves histopathological grading of lung adenocarcinoma. This study presents ANORAK, a deep learning model that uses a multi-order attention mechanism to segment growth patterns in lung adenocarcinoma (LUAD) from hematoxylin and eosin (H&E)-stained slides. ANORAK was trained on data from 49 whole-slide images (WSIs) in the TRACERx 100 cohort and validated on 5,540 WSIs from 1,372 LUAD tumors across four independent cohorts. The model enables precise mapping of growth patterns at the pixel level, facilitating automated grading and analysis of morphological intratumoral heterogeneity. ANORAK demonstrated strong performance in segmenting growth patterns, with higher accuracy than existing methods such as attention U-Net, DeepLabV3+, DANet, and MedT. The model achieved a high correlation with pathologists' estimates for the solid pattern, and showed moderate agreement with pathologists for other patterns. AI grading was found to be an independent prognostic indicator for stage I tumors, with improved survival outcomes compared to pathologists' grading. The model also assisted pathologists in challenging scenarios, such as tumors with highly diversified growth patterns, lepidic-predominant or acinar-predominant tumors, and tumors with high-grade patterns. AI grading consistently outperformed manual grading in these cases, particularly in predicting DFS. Additionally, the model provided insights into the spatial arrangement and morphological features of acinar islands, which may reflect histological transitions and prognosis. The study highlights the potential of AI in improving the accuracy and efficiency of histopathological grading of LUAD. ANORAK provides a robust and automated method for analyzing growth patterns, which can enhance risk stratification and outcome prediction in clinical practice. The model's ability to capture spatial and morphological heterogeneity makes it a valuable tool for understanding the complex biology of LUAD. However, the study acknowledges limitations, including potential errors in segmentation and the need for further validation in larger cohorts. Overall, ANORAK represents a significant advancement in the application of AI to histopathological analysis of LUAD.The artificial intelligence-based model ANORAK improves histopathological grading of lung adenocarcinoma. This study presents ANORAK, a deep learning model that uses a multi-order attention mechanism to segment growth patterns in lung adenocarcinoma (LUAD) from hematoxylin and eosin (H&E)-stained slides. ANORAK was trained on data from 49 whole-slide images (WSIs) in the TRACERx 100 cohort and validated on 5,540 WSIs from 1,372 LUAD tumors across four independent cohorts. The model enables precise mapping of growth patterns at the pixel level, facilitating automated grading and analysis of morphological intratumoral heterogeneity. ANORAK demonstrated strong performance in segmenting growth patterns, with higher accuracy than existing methods such as attention U-Net, DeepLabV3+, DANet, and MedT. The model achieved a high correlation with pathologists' estimates for the solid pattern, and showed moderate agreement with pathologists for other patterns. AI grading was found to be an independent prognostic indicator for stage I tumors, with improved survival outcomes compared to pathologists' grading. The model also assisted pathologists in challenging scenarios, such as tumors with highly diversified growth patterns, lepidic-predominant or acinar-predominant tumors, and tumors with high-grade patterns. AI grading consistently outperformed manual grading in these cases, particularly in predicting DFS. Additionally, the model provided insights into the spatial arrangement and morphological features of acinar islands, which may reflect histological transitions and prognosis. The study highlights the potential of AI in improving the accuracy and efficiency of histopathological grading of LUAD. ANORAK provides a robust and automated method for analyzing growth patterns, which can enhance risk stratification and outcome prediction in clinical practice. The model's ability to capture spatial and morphological heterogeneity makes it a valuable tool for understanding the complex biology of LUAD. However, the study acknowledges limitations, including potential errors in segmentation and the need for further validation in larger cohorts. Overall, ANORAK represents a significant advancement in the application of AI to histopathological analysis of LUAD.