Background: Non-small cell lung cancer (NSCLC) is the most prevalent form of lung cancer, with lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) being the predominant subtypes. Given the significant differences in treatment approaches and clinical outcomes between LUAD and LUSC, accurate identification of these pathological subtypes prior to treatment initiation is crucial. Methods: This study aimed to enhance the diagnostic accuracy for LUAD and LUSC by integrating a machine learning artificial intelligence (AI) model with the expertise of pathologists. The AI model was trained and validated on a cohort of 10,693 and 5,571 cases from the OrigiMed database, each confirmed by at least three pathologists. Targeted sequencing of 450 cancer-related genes was performed. Results: Analysis of the validation set demonstrated an overall accuracy of 89.2% across 28 tumor types, with individual accuracies ranging from 22.2% to 100%. For the top one prediction, the AI model achieved an accuracy of 82.7% for LUAD and 67.9% for LUSC. However, when considering the top three predictions, the accuracy significantly increased to 92.1% for LUAD and 93.0% for LUSC. To further validate the diagnostic capability of the AI model, a large sample of 4,531 LUAD cases and 207 LUSC cases from 4,502 patients was collected for comparative analysis with pathologists' diagnoses. The results indicated that 87.9% of LUAD cases and 83.6% of LUSC cases were consistent with the AI model's diagnosis, while 4.9% of LUAD and 9.7% of LUSC cases showed discrepancies. To investigate these inconsistencies, a subset of 30 LUAD cases and 16 LUSC cases was selected for re-evaluation by independent pathologists. Four poorly differentiated cases from the discrepancy group were ultimately diagnosed as 2 LUADs and 2 LUSCs, supporting the initial AI diagnosis. For instance, Patient 40 was initially diagnosed with LUSC; however, the AI system identified the pathology as LUAD, which may be attributed to the detection of the KIF5B-RET fusion. This diagnosis was subsequently corroborated by an additional pathologist who observed gland-like structures and ambiguous solid mass formations within the tumor tissue, consistent with LUAD histopathology. Conclusions: In summary, we have presented a machine learning AI model that leverages next-generation sequencing (NGS) technology for the auxiliary diagnosis of cancer. By applying this AI model to diagnose LUAD and LUSC and comparing it with the diagnoses made by human pathologists, we have demonstrated that the integration of pathological diagnosis and AI machine learning can significantly enhance and even surpass the diagnostic capabilities of human physicians. This advancement holds great potential in improving the accuracy of tumor diagnosis and the precision of treatment.