This study looked at how well computer models could tell apart two specific types of lung cancer using CT scan images. The researchers examined data from 58 patients with solitary pulmonary invasive mucinous adenocarcinoma and 105 patients with pulmonary invasive non-mucinous adenocarcinoma. They compared three different prediction approaches to see which worked best for diagnosis.
The model that combined both CT morphological features and histogram features achieved the highest accuracy. It had an area under the curve of 0.845, which was higher than the separate models. The separate morphological model scored 0.754, while the separate histogram model scored 0.820.
The authors suggest this combined approach could be a practical tool for non-invasive diagnosis. However, because this was a retrospective review of a limited number of patients, the findings need further testing. Readers should understand this is not a new treatment but a potential diagnostic aid that requires more validation.