Machine learning-based histopathological features of histological slides and clinical characteristics as a novel prognostic indicator in diffuse large B-cell lymphoma.

OBJECTIVE

This study developed and validated a deep learning model based on clinical and histopathological features for predicting the outcomes of diffuse large B-cell lymphoma (DLBCL).

METHODS

This study analyzed 194 whole slide images from 194 patients with DLBCL. Clinical characteristics and histopathological features of hematoxylin-eosin-stained sections were extracted using CellProfiler. These features were analyzed and validated. The prognostic value of these features was evaluated by Cox regression analysis, the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA).

RESULTS

A total of 1120 digital features were extracted using a fully automated process. Harrell's concordance index of the clinicopathologic nomogram was significantly higher than that of the Pathomics score based nomogram (0.791 vs. 0.750). The clinicopathologic nomogram had higher accuracy in predicting overall survival (OS). The AUC of the Pathomics score based nomogram for 1-year and 2-year OS was significantly higher than that of the clinicopathologic nomogram (1-year OS: 0.892 vs. 0.810; 2-year OS: 0.824 vs. 0.764). Nonetheless, the clinicopathologic nomogram had a stronger ability to predict 3-year OS than the simple nomogram (AUC: 0.812 vs. 0.759). DCA confirmed that the clinicopathologic nomogram was a better predictor of long-term OS, improving clinical decision-making.

CONCLUSION

The nomogram based on clinical and histopathological features is a novel, non-invasive, and convenient method to predict OS in patients with DLBCL and can potentially predict responses to treatment.