by Christos Evangelou, MSc, PhD – Medical Writer and Editor

In a new study, researchers at Henan University and Fourth Military Medical University in China developed a new artificial intelligence (AI)-driven pathomics approach to enhance the ability of CDKN2A to predict prognosis in patients with head and neck squamous cell carcinoma (HNSCC).

Validation studies showed that this novel method, which extracts valuable information from pathology images to predict gene expression levels, could improve clinical decision-making by enhancing the prognostic value of CDKN2A.1

“The assessment of CDKN2A expression can significantly influence treatment decisions and patient outcomes by providing a more accurate prognosis. Higher CDKN2A expression is associated with improved survival rates, guiding clinicians to tailor treatment plans more effectively. This could lead to better-informed decisions on the aggressiveness of interventions and the potential use of targeted therapies, ultimately improving patient survival and quality of life,”

said Junpeng Luo, PhD, the corresponding author of the study.

The report was published in the Journal of Cellular and Molecular Medicine.

Need for Accessible and Accurate Prognostic Tools in HNSCC

HNSCC accounts for approximately 90% of all head and neck tumors globally. Despite advancements in surgical techniques and adjuvant therapies, the overall prognosis for HNSCC patients remains poor, with low 5-year survival rates.1

The CDKN2A gene, known for its role in cell cycle regulation, has emerged as a potential prognostic marker in various cancers, including HNSCC. However, current methods for assessing CDKN2A expression are limited by factors such as cost, invasiveness, and reliability. This study aimed to address these limitations by developing a novel, AI-driven approach to analyze CDKN2A expression from pathology images, potentially offering a more accessible and accurate prognostic tool.¹

“The poor prognosis of HNSCC inspired the exploration of novel prognostic markers. Leveraging AI-driven pathomics offers a promising, non-invasive solution to enhance prognosis prediction by precisely correlating CDKN2A expression with patient outcomes,”

noted Dr. Luo.

Approach: Combining High-Throughput Data and AI

The researchers analyzed data from 475 patients with HNSCC from The Cancer Genome Atlas (TCGA) database. These cases were selected to ensure comprehensive clinical and transcriptomic data availability. The team then categorized patients into CDKN2A-high and CDKN2A-low expression groups based on a predetermined expression threshold.¹

The team developed an AI-driven pathomics framework by analyzing 271 cases with available pathology slides, from which they extracted 465 distinct features. These features were used to construct a Gradient Boosting Machine model, designed to compute pathomics scores and predict CDKN2A expression levels.

The researchers divided their dataset into training (70%) and test (30%) sets to validate the performance of the model. They employed various statistical techniques, including Receiver Operating Characteristic (ROC) analysis, Kaplan-Meier survival analysis, and Cox regression analysis, to validate their findings and assess the prognostic value of CDKN2A expression and the pathomics model.¹

Commenting on the novelty of this approach, Dr. Luo said:

“Our AI-driven pathomics framework improves upon traditional methods by providing a non-invasive, highly accurate, and reproducible analysis of CDKN2A expression from pathology images. Traditional methods often rely on invasive biopsies and can be inconsistent. Our approach utilizes machine learning algorithms to extract and analyze extensive pathologic image features, offering a more precise and comprehensive assessment of gene expression and its correlation with clinical outcomes.”

CDKN2A Expression and Patient Outcomes

Patients with higher CDKN2A expression demonstrated significantly longer median overall survival than those with lower expression (66.73 months versus 42.97 months, P = 0.013).1 This association held true in both univariate and multivariate analyses, confirming the role of CDKN2A as an independent prognostic indicator in patients with HNSCC.

Moreover, the study showed that the AI-powered pathomics model developed by the team had strong predictive capability. With area under the curve (AUC) values of 0.806 for the training set and 0.710 for the validation set, the model demonstrated its ability to accurately predict CDKN2A expression levels from pathology images alone. This performance was further validated through tissue microarray analysis, which corroborated the model’s predictive capacity.¹

According to Dr. Luo, these findings suggest that their AI-powered pathomics model could enhance personalized medicine in oncology.

“By providing a detailed molecular understanding of HNSCC through CDKN2A expression analysis, clinicians can develop more individualized treatment plans. This approach not only improves prognostic accuracy but also paves the way for more targeted and effective therapies, aligning with the goals of precision medicine to optimize patient care,”

he said.

Molecular Insights: Beyond Gene Expression

After confirming the prognostic value of CDKN2A in patients with HNSCC, the researcher went one step further to obtain molecular insights using gene set enrichment analysis (GSEA). High pathomics scores were strongly associated with the activation of cell cycle and G2M checkpoint pathways, highlighting a distinct proliferation signature. Conversely, low pathomics scores correlated with pathways such as JAK/STAT signaling and epithelial-mesenchymal transition (EMT), suggesting different oncogenic mechanisms at play.

Interestingly, the team also found that the high pathomics score group exhibited a lower frequency of TP53 mutations, a finding of particular note given the association of TP53 mutations with poor prognosis in HNSCC.

Looking Ahead

The researchers acknowledge the potential for selection bias inherent in their retrospective design. They also emphasize the need for prospective, multicenter studies to further validate and refine their prognostic models.

“Following the results of this study, the next steps in research include validating the AI-driven pathomics model in larger, multicentric prospective studies to confirm its robustness and generalizability,”

noted Dr. Luo.

He added that exploring the integration of broader omics data, such as genomics and proteomics, can further refine the model’s predictive accuracy and provide deeper insights into the molecular mechanisms of HNSCC.

“These advancements will support the continued development of personalized therapeutic strategies in oncology,”

he said.

Regarding the integration of AI-driven pathomics into clinical workflows, Dr. Luo said that although AI-driven pathomics could offer precise, cost-effective, and non-invasive methods for cancer detection and prognostication, potential challenges include the need for robust IT infrastructure, training for medical professionals, and ensuring data security and patient privacy.

“Collaboration with regulatory bodies will be crucial to facilitate the adoption of these advanced technologies in clinical practice,”

he said.

The study received financial support from the Provincial Ministerial Co-Construction Key Project of the Henan Provincial Medical Science and Technology Public Relations Program.

References

1. Wang Y, Zhou C, Li T, Luo J. Prognostic value of CDKN2A in head and neck squamous cell carcinoma via pathomics and machine learning. J Cell Mol Med. 2024;28(9):e18394. doi:10.1111/jcmm.18394