Optimized grade group for reporting prostate cancer grade in systematic and MRI-targeted biopsies

Mitigating bias in prostate cancer diagnosis using synthetic data for improved AI driven Gleason grading

Prostate cancer (PCa) is a leading cause of cancer-related mortality in men, with Gleason grading critical for prognosis and treatment decisions. Machine learning (ML) models offer potential for automated grading but are limited by dataset biases, staining variability, and data scarcity, reducing their generalizability. This study employs generative adversarial networks (GANs) to generate high-quality synthetic histopathological images to address these challenges. A conditional GAN (dcGAN) was developed and validated using expert pathologist review and Spatial Heterogeneous Recurrence Quantification Analysis (SHRQA), achieving 80% diagnostic quality approval. A convolutional neural network (EfficientNet) was trained on original and synthetic images and validated across TCGA, PANDA Challenge, and MAST trial datasets. Integrating synthetic images improved classification accuracy for Gleason 3 (26%, p = 0.0010), Gleason 4 (15%, p = 0.0274), and Gleason 5 (32%, p < 0.0001), with sensitivity and specificity reaching 81% and 92%, respectively. This study demonstrates that synthetic data significantly enhances ML-based Gleason grading accuracy and improves reproducibility, providing a scalable AI-driven solution for precision oncology.

Does Overgrading on Targeted Biopsy of Magnetic Resonance Imaging-visible Lesions in Prostate Cancer Lead to Overtreatment?

BACKGROUND AND OBJECTIVE

Targeted biopsy of the index prostate cancer (PCa) lesion on multiparametric magnetic resonance imaging (MRI) is effective in reducing the risk of overdiagnosis of indolent PCa. However, it remains to be determined whether MRI-targeted biopsy can lead to a stage shift via overgrading of the index lesion by focusing only on the highest-grade component, and to a subsequent risk of overtreatment. Our aim was to assess whether overgrading on MRI-targeted biopsy may lead to overtreatment, using radical prostatectomy (RP) specimens as the reference standard.

METHODS

Patients with clinically localized PCa who had positive MRI findings (Prostate Imaging-Reporting and Data System [PI-RADS] score ≥3) and Gleason grade group (GG) ≥2 disease detected on MRI-targeted biopsy were retrospectively identified from a prospectively maintained database that records all RP procedures from eight referral centers. Biopsy grade was defined as the highest grade detected. Downgrading was defined as lower GG for the RP specimen than for MRI-targeted biopsy. Overtreatment was defined as downgrading to RP GG 1 for cases with GG ≥2 on biopsy, or to RP low-burden GG 2 for cases with GG ≥3 on biopsy.

KEY FINDINGS AND LIMITATIONS

We included 1020 consecutive biopsy-naïve patients with GG ≥2 PCa on MRI-targeted biopsy in the study. Pathological analysis of RP specimens showed downgrading in 178 patients (17%). The transperineal biopsy route was significantly associated with a lower risk of downgrading (odds ratio 0.364, 95% confidence interval 0.142-0.814; p = 0.022). Among 555 patients with GG 2 on targeted biopsy, only 18 (3.2%) were downgraded to GG 1 on RP. Among 465 patients with GG ≥3 on targeted biopsy, three (0.6%) were downgraded to GG 1 and seven were downgraded to low-burden GG 2 on RP. The overall risk of overtreatment due to targeted biopsy was 2.7% (28/1020).

CONCLUSIONS AND CLINICAL IMPLICATIONS

Our multicenter study revealed no strong evidence that targeted biopsy results could lead to a high risk of overtreatment.

A novel biopsy scheme for prostate cancer: targeted and regional systematic biopsy

PURPOSE

To explore a novel biopsy scheme for prostate cancer (PCa), and test the detection rate and pathological agreement of standard systematic (SB) + targeted (TB) biopsy and novel biopsy scheme.

METHODS

Positive needles were collected from 194 patients who underwent SB + TB (STB) followed by radical prostatectomy (RP). Our novel biopsy scheme, targeted and regional systematic biopsy (TrSB) was defined as TB + regional SB (4 SB-needles closest to the TB-needles). The McNemar test was utilized to compare the detection rate performance for clinical significant PCa (csPCa) and clinical insignificant PCa (ciPCa). Moreover, the accuracy, positive predictive value (PPV) and negative predictive value (NPV) were investigated. The agreement between the different biopsy schemes grade group (GG) and RP GG were assessed. The concordance between the biopsy and the RP GG was evaluated using weighted κ coefficient analyses.

RESULTS

In this study, the overall detection rate for csPCa was 83.5% (162 of 194) when SB and TB were combined. TrSB showed better NPV than TB (97.0% vs. 74.4%). Comparing to STB, the TB-detection rate of csPCa had a significant difference (p < 0.01), while TrSB showed no significant difference (p > 0.999). For ciPCa, the overall detection rate was 16.5% (32 of 194). TrSB showed better PPV (96.6% vs. 83.3%) and NPV (97.6% vs. 92.9%) than TB. Comparing to STB, the detection rate of both schemes showed no significant difference (p = 0.077 and p = 0.375). All three schemes GG showed poor agreement with RP GG (TB: 43.3%, TrSB: 46.4%, STB: 45.9%). Using weighted κ, all three schemes showed no difference (TB: 0.48, TrSB: 0.51, STB: 0.51). In our subgroup analysis (PI-RADS = 4/5, n = 154), all three schemes almost showed no difference (Weighted κ: TB-0.50, TrSB-0.51, STB-0.50).

CONCLUSION

Our novel biopsy scheme TrSB (TB + 4 closest SB needles) may reduce 8 cores of biopsy compared with STB (standard SB + TB), which also showed better csPCa detection rate than TB only, but the same as STB. The pathological agreement between three different biopsy schemes (TB/TrSB/STB) GG and RP GG showed no difference.

Prostate magnetic resonance imaging-targeted biopsy global grade correlates better than highest grade with prostatectomy grade

BACKGROUND

Magnetic resonance imaging (MRI)-targeted prostate biopsy has become an increasingly common method of diagnosing prostate cancer. A previous study from our institution demonstrated that the biopsy global Grade Group (gGG, aggregate GG of all positive cores) and highest Grade Group (hGG in any core) both show substantial concordance with the Grade Group at radical prostatectomy (RPGG) while the discordance predominantly consists of upgrading in gGG and downgrading in hGG. We performed a larger cohort study focused on biopsy cases in which gGG and hGG differ, to determine their relative concordance with RPGG.

METHODS

We conducted a retrospective review of radical prostatectomy specimens with prior MRI-targeted biopsies from our institution between 2016 and 2020. Separate gGG and hGG were assigned to each MRI-targeted lesion. Targeted lesions with different gGG versus hGG were segregated from those with identical gGG and hGG. The concordance of biopsy GG with RPGG was evaluated using κ coefficient analysis.

RESULTS

Of the 489 lesions with MRI-targeted biopsies, 82 (17%) differed in gGG versus hGG. The gGG of 46 (56%), 33 (40%), and 3 (4%) lesions were unchanged, upgraded, and downgraded at radical prostatectomy, respectively (κ= 0.302, weighted κ = 0.334). The hGG of 24 (29%), 9 (11%), and 49 (60%) lesions were unchanged, upgraded, and downgraded at radical prostatectomy, respectively (κ = 0.040, weighted κ = 0.198). When stratified by the biopsy GG, gGG showed the highest concordance in GG2 (61%) and GG3 (54%) lesions. The hGG resulted in substantial downgrading (60%) with less optimal concordance regardless of the biopsy GG. Neither the prebiopsy prostate specific antigen level nor the PI-RADS score was predictive of upgrading of gGG.

CONCLUSIONS

When gGG and hGG differ, gGG method more accurately predicts the RPGG than hGG, particularly in GG2 and GG3 lesions which comprised the majority of targeted lesions.