Background

PSA-based screening has come in for a lot of criticisms due to noticeable false positive results for prediction of prostate cancer (PCa).

Purpose

To investigate whether a machine learning analysis of multi-parametric (mp) MRI can help to improve the predictive performance in PCa.

Design, setting, and participants

Based on a support vector machine classification analysis, we modeled clinical data (age, PSA, DRE, TRUS, PSAD, prostate volume) and mp-MRI findings on 985 men between Jan 2002 and Dec 2009 to predict the probability of PCa. The model was validated on 493 patients treated at a same institution during the same period. Median follow-up in the training and validation cohorts was 40 and 38 mo, respectively.

Measurements

5-yr rate of PCa were measured.

Results and limitations

At 5-yr follow-up period, 34.3% of patients had systemic progression of PCa. By adding mp-MRI findings, the machine learning-based model increased area under curve (Az) from 0.715 to 0.938 (p < 0.001), sensitivity from 27.4% to 83.5% (p < 0.001), positive predictive value (PPV) from 57.7% to 77.4% (p = 0.004), and negative predictive value (NPV) from 71.3% to 91.5% (p < 0.001). Scaled by PSA level, the model had high PPV (93.6%) in patients with PSA > 20 ng/ml, while with intermediate to low PPVs in group of PSA 10-20 ng/ml (64%), PSA 4-10 ng/ml (55.8%) and PSA 0-4 ng/ml (29%). Among all predictors, MR PI-RADS score had the highest Cox hazard ratio (2.112; p < 0.001) associated with time to progression of PCa, followed by PSA (1.435; p < 0.001), age (1.012; p = 0.043).

Discussion

To our knowledge, this is the first prospective study to investigate the predictive role of pre-biopsy mp-MRI for PCa with a long-term follow-up in 1478 consecutive patients. We report that, at 5-yr follow-up period, 34.3% of patients had systemic progression of PCa, and 29.6% was detected within initial 1-yr follow-up, while only 2.5% was detected within the following 4-yr period. In addition, we noted that the machine learning-based model using the input variables of clinical variables and mp-MRI findings (SVM-MRI) shows high sensitivity (83.5%) and specificity (87.8%) to predict PCa statistically more accurate (86.4%) than PSA-based screening program. And importantly, the high negative predictive values of 91.5% made the mp-MRI extraordinarily useful for initial evaluation before an invasive biopsy.

Conclusion

We first investigated the systemic outcome of a crowd underwent PSA-based screening and pre-biopsy mp-MRI, and demonstrated the predictive role of pre-biopsy mp-MRI for PCa by using an advanced machine learning-based approach. Here we answer one important question at the beginning of the paper: (1) mp-MRI coupled with PSA screening program can be used to detect PCa. By the constructed nomogram, the outcome of most patients could be accurately predicted in the first 1-yr follow-up period if they received a pre-biopsy mp-MRI examination, even without invasive TRUS biopsy.

Acknowledgements

No acknowledgement found.

References

No reference found.