Selective identification of polyploid hepatocellular carcinomas with poor prognosis by artificial intelligence-based...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Polyploidy, an abnormal chromosome number, is common in cancer and linked to aggressive disease and poor prognosis.
• Hepatocellular carcinoma (HCC) exhibits polyploidy, making it a prognostic marker, yet diagnosis remains challenging.
• Current diagnostic methods for polyploidy in cancer are not clinically implementable.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based method for assessing polyploidy in HCC using pathological images.
• To evaluate the AI model's ability to detect HCC ploidy and identify prognostic subgroups.

Main Methods:

• Deep learning-based AI image recognition models were trained on hematoxylin and eosin-stained pathological images of HCC.
• Model performance was evaluated using chromosome fluorescence in situ hybridization (FISH) determined ploidy status in 44 HCCs.
• The validated models were tested on independent cohorts of 169 HCCs and a public dataset.

Main Results:

• The AI models accurately assessed HCC ploidy in independent and external datasets.
• AI-driven ploidy determination identified a subset of HCCs with poor prognosis, outperforming genomic analysis.
• AI analysis revealed novel insights into polyploid HCC, including a high prevalence of scirrhous structures.

Conclusions:

• AI-based pathological image analysis offers a novel and effective tool for assessing polyploidy in HCC.
• This approach has the potential to serve as a diagnostic tool for personalized cancer medicine.
• The AI pipeline can identify prognostically significant subgroups within HCC based on ploidy status.