# Histopathologic deep learning model for predicting tumor response to hepatic arterial infusion chemotherapy plus TKIs and ICIs in large hepatocellular carcinoma

## TL;DR

A deep learning model called HAIM was developed to predict how large liver cancer patients respond to a combination therapy of chemotherapy, tyrosine kinase inhibitors, and immune checkpoint inhibitors.

## Contribution

The study introduces HAIM, a novel deep learning model using histopathological images to predict response to triplet therapy in hepatocellular carcinoma.

## Key findings

- HTI group showed significantly better outcomes than HAIC group in ORR, PFS, and OS.
- HAIM achieved AUC scores of 0.778 in predicting HTI response from histopathological images.
- No significant differences in ORR were observed within the HTI group across different BCLC stages.

## Abstract

While triplet therapy (HTI), which combines hepatic arterial infusion chemotherapy (HAIC) with tyrosine kinase inhibitors and immune checkpoint inhibitors, is widely used in the treatment of large hepatocellular carcinoma (HCC), there are few reports about its efficacy versus HAIC, and no reliable methods are available for promptly predicting HTI response.

This study included treatment-naïve patients with large HCCs (> 5 cm in diameter) from two centers between January 2017 and December 2022. Objective response rate (ORR), progression-free survival (PFS), and overall survival (OS) were compared between the HTI and HAIC groups. To efficiently predict HTI response, available pre-treatment H&E-stained biopsy slides of HCC patients were collected to develop deep-learning models.

Compared to group HAIC (n = 97), group HTI (n = 281) showed an ORR (54.45% vs. 21.65%), PFS (median, 10.9 vs. 4.9 months), and OS (median, 25.0 vs. 12.0 months). No significant differences in ORR were observed within the HTI group across different BCLC stages. A deep learning model, termed the Hepatocellular Carcinoma Artificial Intelligence Prediction Model (HAIM), was developed using pathological slides of HTI-treated patients (n = 194). HAIM achieved AUC scores of 0.778 (entire testing set), 0.735 (internal testing set), and 0.853 (external testing set).

Integrating TKIs and ICIs with HAIC significantly improved ORR, PFS, and OS in all stages of large HCCs. HAIM, derived from histopathological images of the biopsy, showed potential clinical aid for predicting HTI response, providing a novel tool for personalized management of HCC.

The online version contains supplementary material available at 10.1186/s40644-025-00885-x.