According to the study authors, an end-to-end deep learning approach to generate CT-based biomarkers may help produce more rapid clinical translation.
A diagnostic/prognostic study published in JAMA Network Open provided a clinically applicable, end-to-end CT–based deep learning approach which was able to determine which patients with stage IV EGFR variant–positive non–small cell lung cancer (NSCLC) are more likely to benefit from EGFR tyrosine kinase inhibitor (TKI) therapy based on disease course.
Specifically, investigators used a state-of-the-art representation learning framework to build a deep learning semantic prognosis signature from pretherapy CT images obtained from these patients. According to the investigators, an end-to-end deep learning approach to generate CT-based biomarkers such as this may help produce more rapid clinical translation.
“Compared with radiomics, the approach used in this study not only avoids human intervention procedures requiring manual region of interest segmentation and predesigning heterogeneous features, but it also appears to be associated with better survival prognostic performance,” wrote the study authors, who were led by Jiangdian Song, PhD. “Finally, the open-access source code provided in this study enables interested readers to reproduce the results of this study.”
From January 1, 2010 to August 1, 2017, a total of 465 patients were enrolled in the study, with follow-up from February 1, 2010 to June 1, 2020. Using a training cohort, a deep learning semantic signature was composed to predict progression-free survival (PFS). The signature was then validated in 2 external validation cohorts and 2 control cohorts, and thereafter compared with the radiomics signature.
The primary end point of the study was PFS, which started at the initiation of therapy up to the date of recurrence, confirmed disease progression, or death.
Overall, 342 patients met the inclusion criteria, including 56 patients with advanced-stage EGFR variant–positive NSCLC and 67 patients with advanced-stage EGFR wild-type NSCLC who received first-line chemotherapy. Of the total study cohort, 145 patients from 2 hospitals (n = 117 and 28) formed a training cohort, and patients from 2 other hospitals comprised 2 external validation cohorts (validation cohort 1: n = 101; and validation cohort 2: n = 96).
Using the deep learning semantic signature, patients predicted to have high risk of rapid progression accounted for 26% of all patients and had a median PFS that was similar to patients with advanced-stage NSCLC who received conventional chemotherapy.
When comparing the median PFS of high-risk patients receiving EGFR-TKI therapy with the chemotherapy cohorts, no significant differences were observed (median PFS, 6.9 vs 4.4 months; P = .08). However, with regard to predicting the tumor progression risk following EGFR-TKI therapy, clinical decisions based on the deep learning semantic signature led to improved survival outcomes compared with those based on radiomics signature across all risk probabilities by the decision curve analysis.
“In this study, a [deep learning]–-based approach was used to identify significant semantic features without requiring tumor margin delineation. Thus, unlike the more labor-intensive radiomics-based image predictors, an end-to-end [deep learning] approach to generate CT-based biomarkers to stratify patients who are more likely to benefit from EGFR-TKI therapy can facilitate more rapid clinical translation,” the authors explained. “Direct comparisons between the image features extracted by [deep learning] networks and those extracted by radiomics are currently lacking. These results also suggest that [deep learning]–-based semantic features achieve better performance than radiomics in predicting the efficacy of EGFR-TKI therapy.”
Importantly, in order to train the model and reduce unnecessary calculations, this study used CT images which only included lung tumors; therefore, screening of CT sections is still necessary. In addition, there is not currently a specific definition of the deep learning semantic features, and the investigators indicated they intend to further explore the image encoding process corresponding to the deep learning semantic features in order to determine the specific image characteristics mirrored by these features.
Song J, Wang L, Norton N, et al. Development and validation of a machine learning model to explore tyrosine kinase inhibitor response in patients with stage IV EGFR variant-positive non-small cell lung cancer. JAMA Network Open. 2020;3(12):e2030442. doi: 10.1001/jamanetworkopen.2020.30442