Abstract 2339: Development of a blood cell analysis-based AI model for detecting early ovarian cancer signals

Ovarian cancer (OC) remains one of the most lethal gynecologic malignancies, primarily due to delayed detection. Blood cell analysis presents a promising non-invasive diagnostic approach, capable of detecting systemic signals associated with early OC. This study focuses on developing and validating an artificial intelligence (AI)-based diagnostic model to detect OC risk signals using comprehensive blood cell data. To minimize confounding factors, samples from male participants, symptomatic controls, drug-treated individuals, and pre-anesthesia collections were excluded. Additionally, samples with missing data were removed. A total of 185 samples, including 85 for training and 100 for testing, were analyzed, comprising OC, benign tumor, and control groups. The model utilized 24 features: 18 directly measured blood cell parameters (via Sysmex) and 6 derived composite indices, including the systemic immune-inflammation index (SII: [NEUT×PLT]/LYM). A Random Forest algorithm, optimized through 5-fold GridSearch cross-validation and soft voting, was trained and validated with balanced accuracy (BA) as the performance metric. The optimal cutoff values were determined based on the training set’s best BA. In the training set, the model achieved a sensitivity of 100.0%, specificity of 92.6%, a positive predictive value (PPV) of 77.3%, and a negative predictive value (NPV) of 100.0%, with an area under the curve (AUC) of 0.987. For the test dataset, the model demonstrated a sensitivity of 75.0%, specificity of 92.9%, PPV of 66.7%, and NPV of 95.1%, with an AUC of 0.923. The platelet-to-lymphocyte ratio (PLT/LYM) and SII emerged as the most influential features, contributing significantly to model performance. The specificity for control samples, excluding benign tumors, was 1 in the training set, and validation confirmed a specificity of 0.984, demonstrating clear separation between the OC and control groups. Composite indices further improved the model’s reliability in detecting OC risk signals. This AI-driven blood cell analysis model exhibited strong performance in identifying early ovarian cancer signals, achieving high sensitivity and specificity. By leveraging non-invasive biomarkers, it holds significant promise for clinical implementation in OC screening. Future efforts will focus on expanding the diversity of analyzed samples to better distinguish cancer-specific signals from those associated with other diseases and incorporating molecular markers to enhance the accuracy and precision of the diagnostic algorithm. Citation Format: Eunyong Ahn, Sungmin Park, Sarah Kim, Se Ik Kim, Hyejin Lee, Heeyeon Kim, Seong Eun Kang, Ji Won Park, TaeJin Ahn, Yong-Sang Song. Development of a blood cell analysis-based AI model for detecting early ovarian cancer signals [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2339.