Artificial Intelligence in Predictive Oncology: A Clinical Study Using Machine Learning for Cancer Detection

Abstract

Background: Cancer remains a leading global health challenge, with early and accurate detection being critical for improving patient outcomes. Traditional diagnostic methods such as biopsy and advanced imaging have limitations in accessibility, cost, invasiveness, and observer variability. The recent emergence of artificial intelligence (AI), specifically machine learning (ML) and deep learning (DL), offers promising advances in cancer detection by efficiently analysing multimodal data and uncovering subtle, clinically relevant patterns that may elude conventional approaches. The aim of this study is to evaluate and compare artificial intelligence-based predictive models for cancer detection across breast, lung, and colorectal cancers using multimodal data integration.

Methodology: This study retrospectively analysed a synthetic dataset of 600 patients representing breast, lung, and colorectal cancer. Patient profiles included demographics, laboratory biomarkers, imaging attributes, and genetic mutations. Predictive models—logistic regression, random forest, support vector machine (SVM), and deep neural network (DNN)—were trained and rigorously evaluated using stratified sampling, 10-fold cross-validation, and grid or iterative hyperparameter tuning. Model performance was assessed via metrics including accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). SHAP (Shapley Additive exPlanations) was used to ensure model interpretability and clinical relevance, identifying the most influential predictors. Statistical analyses (McNemar's test, paired t-test, DeLong's test) were applied to compare model performance, with significance set at p < 0.05.

Results: The DNN outperformed all other models, achieving an accuracy of 94.8%, precision of 94.2%, recall of 93.5%, F1-score of 93.8%, and AUC of 0.96. Tumour size, smoking history, and cancer-associated genetic mutations (BRCA for breast, KRAS for colorectal) emerged as the strongest predictors for cancer detection. The DNN demonstrated strong recall for all cancer types (breast: 92%, lung: 95%, colorectal: 94%), significantly reducing false negatives and false positives. Statistical tests confirmed that the DNN’s performance advantages were significant compared to baseline and ensemble approaches.

Conclusion: AI-based predictive modelling, particularly deep learning, substantially improves the accuracy and reliability of cancer detection. Integrating imaging, clinical, and genetic biomarkers within a unified framework enhances early diagnosis and personalised oncology, supporting radiology and pathology workflows. Explainable AI methods strengthen clinical trust and offer transparency in decision-making. Implementation of these models can potentially reduce diagnostic delays and democratise access to expertise, especially in resource-limited settings. Future research should focus on real-world validation, multi-centre trials, and ethical assessments to realise AI’s transformative impact in precision oncology.