Deep learning models predicted survival outcomes better than a standard clinical model by analyzing tumor scans taken from patients with lung cancer, according to a study published in Clinical Cancer Research.
Lung cancer is among the most common cancers worldwide. A majority of patients are diagnosed with non-small cell lung cancer (NSCLC) and have a five-year survival rate of 18%, wrote Hugo Aerts, PhD, director of the Computational and Bioinformatics Laboratory, Dana-Farber Cancer Institute and Brigham and Women’s Hospital in Boston, and colleagues. While modern advances have cut into cancer mortality rates, that drop in deaths hasn’t been as prevalent for lung cancer patients.
The team created multiple deep learning models using imageNet, a neural network that identifies ordinary objects from relevant features. Aerts et al. trained their models using serial CT scans of 179 patients with stage III NSCLC who had received chemoradiation treatment. Each patient produced up to four images taken before treatment and at one, three and six month follow-up for a total of 581 images.
Each model was tested on two datasets, a training set of 581 images and an independent validation dataset of 178 images taken from 89 patients with NSCLC who were treated with chemoradiation and surgery.
"Radiology scans are captured routinely from lung cancer patients during follow-up examinations and are already digitized data forms, making them ideal for artificial intelligence applications," Aerts said in a prepared statement. "Deep-learning models that quantitatively track changes in lesions over time may help clinicians tailor treatment plans for individual patients and help stratify patients into different risk groups for clinical trials."
Aerts and colleagues compared their deep learning models to the clinical standard which uses stage, gender, age, tumor grade, performance, smoking status and clinical tumor size. The deep learning model improved with each follow-up scan, predicting two-year survival based on pretreatment scans with an area under the curve score of 0.58. That reached 0.74 after adding all available follow-up scans.
Additionally, the models’ stratified patients into mortality-risk groups, which were highly associated with overall survival. Patients deemed low-risk for mortality had a six-fold improved overall survival compared with those the model considered high risk.
Compared to the clinical model, the deep learning methods were more efficient in predicting metastasis, progression and local regional recurrence.
“Our research demonstrates that deep-learning models integrating routine imaging scans obtained at multiple time points can improve predictions of survival and cancer-specific outcomes for lung cancer," said Aerts. "By comparison, a standard clinical model relying on stage, gender, age, tumor grade, performance, smoking status, and tumor size could not reliably predict two-year survival or treatment response."