Predicting High-Risk Prostate Cancer Using Machine Learning Methods
Problem
PSA-based prostate cancer screening produces both false positives and false negatives, making it hard to identify which patients have high-risk prostate cancer.
Why it matters
Missed high-risk cancers can be fatal while false alarms lead to unnecessary invasive treatment.
What the AI did
Built a machine learning pipeline using standard scaling, SVMSMOTE sampling for imbalanced data, and AdaBoost to classify patients as no cancer, low-risk, or high-risk.
Results
91.5% accuracy predicting high-risk prostate cancer on a 35,875-patient dataset
Impact
Demonstrated on retrospective screening-trial data; 52 citations; no stated clinical deployment.
Record
| Venue | Data |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Henry Barlow, Shunqi Mao, Matloob Khushi |
| Citations | 55 (OpenAlex, at import time) |
| References | 29 |
| Model or method | AdaBoost with SVMSMOTE sampling and standard scaling |
| Organization | National Cancer Institute (dataset source, PLCO trial) |
| Primary source | https://doi.org/10.3390/data4030129 |
| Open access | gold |
| Keywords | machine learning, prostate cancer, PSA screening, AdaBoost, imbalanced data |
Importance rating
major — millions of lives, an industry or a field
PSA screening errors cause missed high-risk prostate cancer and unnecessary treatment; abstract notes negative consequences but gives no population scale.