Abstract 951P
Background
Hepatocellular carcinoma (HCC) is amongst the leading causes of cancer-related death in the world, representing ∼90% of primary liver cancers. Currently, liver transplantation remains the best treatment for cirrhotic patients with early-stage HCC, however tumor recurrence following liver transplant is observed in 15-20% of cases, which correlates with poor survivorship. Moreover, there are currently no reliable histological markers of relapse-free survival in HCC patients following liver transplant, which is critical in predicting patient prognosis.
Methods
298 whole-slide images stained with haematoxylin/eosin from HCC patients treated by liver transplant at Cleveland Clinic were collected and analyzed as a discovery cohort. Follow-up information, clinical and biological data and histological grades were also collected. A deep learning (DL) model was trained to predict relapse-free survival from histological slides. Repeated (five splits with five folds each) cross-validation splits were used to compute robust estimates of C-index. Results were compared to those obtained by fitting a Cox proportional hazard model on clinical, biological, and pathological variables.
Results
The deep learning model was predictive of recurrence among transplanted patients both in the whole cohort and in subgroups of patients who received or not loco-regional therapy prior to transplant. Results were comparable to those obtained from the Cox model incorporating clinical, biological, and pathological features. Notably, combination of both models significantly increased the C-index in the whole cohort and in the subgroup of patients who did not receive locoregional therapy, demonstrating an independent predictive value of the deep learning model. Table: 951P
| C-index on the whole cohort (N=298) (standard deviation) | C-index on patients with loco-regional therapy (N=145) (s.d.) | C-index on patients without loco-regional therapy (N=153) (s.d.) | |
| Cox model | 0.73 (0.09) | 0.70 (0.11) | 0.75 (0.20) |
| DL model | 0.71 (0.10) | 0.64 (0.13) | 0.81 (0.13) |
| Cox model + DL Model | 0.75 (0.10) | 0.67 (0.15) | 0.85 (0.10) |
Conclusions
Our study demonstrates the prognostic power of deep learning applied to histology slides to predict recurrence of patients with HCC following liver transplantation.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
B. Schmauch: Financial Interests, Institutional, Full or part-time Employment: Owkin; Financial Interests, Institutional, Stocks/Shares: Owkin. P. Sin-Chan: Financial Interests, Institutional, Full or part-time Employment: Owkin. All other authors have declared no conflicts of interest.