Radiogenomics for predicting p53 status, PD-L1 expression, and prognosis with machine learning in pancreatic cancer

Abstract

Background: Radiogenomics is an emerging field that integrates "Radiomics" and "Genomics". In the current study, we aimed to predict the genetic information of pancreatic tumours in a simple, inexpensive, and non-invasive manner, using cancer imaging analysis and radiogenomics. We focused on p53 mutations, which are highly implicated in pancreatic ductal adenocarcinoma (PDAC), and PD-L1, a biomarker for immune checkpoint inhibitor-based therapies.

Methods: Overall, 107 patients diagnosed with PDAC were retrospectively examined. The relationship between p53 mutations as well as PD-L1 abnormal expression and clinicopathological factors was investigated using immunohistochemistry. Imaging features (IFs) were extracted from CT scans and were used to create prediction models of p53 and PD-L1 status.

Results: We found that p53 and PD-L1 are significant independent prognostic factors (P = 0.008, 0.013, respectively). The area under the curve for p53 and PD-L1 predictive models was 0.795 and 0.683, respectively. Radiogenomics-predicted p53 mutations were significantly associated with poor prognosis (P = 0.015), whereas the predicted abnormal expression of PD-L1 was not significant (P = 0.096).

Conclusions: Radiogenomics could predict p53 mutations and in turn the prognosis of PDAC patients. Hence, prediction of genetic information using radiogenomic analysis may aid in the development of precision medicine.

Fig. 1. Immunohistochemistry of p53 and PD-L1 in PDAC.

Typical immunohistochemical staining pattern of p53 (a–c) and PD-L1 (d, e). a Normal staining pattern of nuclei in tumour adjacent pancreatic tissue and “negative” staining pattern in PDAC for p53 in IHC. b Abnormal staining pattern in PDAC; nuclear accumulation of p53 protein was observed in IHC, which was defined as “Positive” indicating mutated p53. C, Absence of p53 in PDAC, which was also defined as “Positive”. Example of typical immunohistochemical “Positive” and “Negative” staining pattern of PD-L1, respectively (d, e). Image magnification of ×400.

Fig. 2. Machine learning processing was summarised.

Total 2,074 IFs extracted from two- phase CT images. Predictive models for p53 and PD-L1 were constructed with machine learning from the IFs. The results were visualised and interpreted in AUC plots and Kaplan–Meier plots.

Fig. 3. Kaplan-Meier plots of p53 and PD-L1 status by IHC compared with status by machine-learning, and ROC curve.

Kaplan–Meier plots for patients with PDAC demonstrating prognostic influence for the “real” status (positive/negative) of p53 in IHC (a), PD-L1 in IHC (b) and “predicted” status of p53 (c), PD-L1 (d) with reference to overall survival. ROC curve was constructed by “real” status and “predicted” status of p53 (e) and PD-L1 (f), respectively. “predicted” status was calculated with machine learning and 1037 IFs extracted from CT.