Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge

Artificial intelligence (AI) has shown promise for diagnosing prostate cancer in biopsies. However, results have been limited to individual studies, lacking validation in multinational settings. Competitions have been shown to be accelerators for medical imaging innovations, but their impact is hindered by lack of reproducibility and independent validation. With this in mind, we organized the PANDA challenge—the largest histopathology competition to date, joined by 1,290 developers—to catalyze development of reproducible AI algorithms for Gleason grading using 10,616 digitized prostate biopsies. We validated that a diverse set of submitted algorithms reached pathologist-level performance on independent cross-continental cohorts, fully blinded to the algorithm developers. On United States and European external validation sets, the algorithms achieved agreements of 0.862 (quadratically weighted κ, 95% confidence interval (CI), 0.840–0.884) and 0.868 (95% CI, 0.835–0.900) with expert uropathologists. Successful generalization across different patient populations, laboratories and reference standards, achieved by a variety of algorithmic approaches, warrants evaluating AI-based Gleason grading in prospective clinical trials.

Single-cell analysis of human primary prostate cancer reveals the heterogeneity of tumor-associated epithelial cell states

Prostate cancer is the second most common malignancy in men worldwide and consists of a mixture of tumor and non-tumor cell types. To characterize the prostate cancer tumor microenvironment, we perform single-cell RNA-sequencing on prostate biopsies, prostatectomy specimens, and patient-derived organoids from localized prostate cancer patients. We uncover heterogeneous cellular states in prostate epithelial cells marked by high androgen signaling states that are enriched in prostate cancer and identify a population of tumor-associated club cells that may be associated with prostate carcinogenesis. ERG-negative tumor cells, compared to ERG-positive cells, demonstrate shared heterogeneity with surrounding luminal epithelial cells and appear to give rise to common tumor microenvironment responses. Finally, we show that prostate epithelial organoids harbor tumor-associated epithelial cell states and are enriched with distinct cell types and states from their parent tissues. Our results provide diagnostically relevant insights and advance our understanding of the cellular states associated with prostate carcinogenesis.

The best prostate biopsy sampling system—fusion and systematic biopsy: A single center experience

Background: Prostate cancer is the second most commonly diagnosed cancer in men. The diagnostic accuracy in prostate cancer can be increased by employing a preliminary multiparametric MRI followed by a fusion-targeted biopsy. Methods: To compare the diagnostic accuracy of fusion-targeted biopsy with the standard systematic biopsy in prostate cancer patients, we enrolled 139 patients on which we performed 139 prostate biopsies consisting of three targeted samples followed by 12 regular systematic samples. Based on histology, we analyzed the diagnostic performance of the two methods. Results: Both methods were equally good at detecting clinically significant cancer (83.3%, 50/60), while systematic biopsy detected more clinically insignificant cancers. However, the best diagnostic performance is obtained by combining the two methods. Conclusion: The two methods are best seen as synergistic, and the addition of fusion biopsy can be used to detect more clinically significant prostate cancers than systematic biopsy alone.

The Initial Application of MRI Based Prostate Volume-Adjusted Prostate-Specific Antigen In The Diagonosis of MRI-Positive Prostate Cancer Patients

Purpose: To identify the value of prostate-specific antigen density (PSAD) and prostate-specific antigen density of the transition zone (PSADTZ) in improving the sensitivity and specificity of the prostate multiparameter magnetic resonance imaging (mp-MRI), for the purpose of predicting prostate cancer (PCa) and grade reclassification in men with prostate-specific antigen (PSA) between 4 and 20 ng/mL to reduce unnecessary prostate biopsies. Patients and Methods: Between 2018 and 2020, we retrospectively identified 283 consecutive men in Shanghai Jiao Tong University Affiliated Sixth People’s Hospital who had mp-MRI and PSA test within 3 months before prostate biopsies. Total prostate volume (TPV) and transition zone volume (TZV) were measured on mp-MRI. PSA, PSAD, and PSADTZ were compared to improve the sensitivity and specificity of positive biopsy cores and pathological stage by univariate analyses and through the receiver operating curve (ROC). We were focused primarily on the MRI-positive patients with PSA levels of 4-20ng/ml who were most likely subjected to unnecessary repeated prostate biopsies. Results: Of the 283 patients, 138 (48.8%) had PCa and in 145 (51.2%) a benign prostate disease was diagnosed. PSA, PSAD, and PSADTZ were significantly related to biopsy, and equally able to predict higher pathological stage. The receiver operating curve (AUC) for predicting the presence of PCa in all patients was 58.06 for PSA, 72.13 for PSAD and 78.28 for PSADTZ. In addition, the AUC for predicting higher pathological stage in PCa patients was 65.71 for PSA, 65.46 for PSAD and 69.81 for PSADTZ. For 228 MRI-positive patients, the AUC for predicting the presence of PCa was 61.31 for PSA, 74.00 for PSAD and 80.13 for PSADTZ. No difference among the PSA, PSAD, and PSADTZ was found in 55 MRI-negative patients. Conclusion: The determination of PSADTZ had higher diagnostic accuracy for PCa than that based on PSA or PSAD. For MRI-positive patients, PSADTZ promote a more effective and simple method for PCa detection, and may be useful for decreasing the burden of surveillance prostate biopsies.

Clinical strategy of repeat biopsy in patients with atypical small acinar proliferation (ASAP)

Atypical small acinar proliferation (ASAP) occurs in approximately 5% of prostate biopsies. Approximately 30–40% of patients with ASAP have biopsy detectable prostate cancer (PCa) within 5 years. Current guidelines recommend a repeat biopsy within 3–6 months after the initial diagnosis. The aim of the present study was to examine the association between ASAP and subsequent diagnosis of clinically significant PCa (csPCa). The need for immediate repeat biopsy was also evaluated. We identified 212 patients with an ASAP diagnosis on their first biopsy at our institution between February 2006 and March 2018. Of these patients, 102 (48.1%) had at least one follow-up biopsy. Clinicopathologic features including rates of subsequent PCa and csPCa were assessed. Thirty-five patients subsequently underwent radical prostatectomy (RP). Their pathologic results were reviewed. csPCa was defined as the presence of Gleason score (GS) ≥ 3 + 4 in ≥ 1 biopsy core. Adverse pathology (AP) was defined as high-grade (primary Gleason pattern ≥ 4) or non-organ-confined disease (pT3/N1) after RP. Of 102 patients, 87 (85.3%), 13 (12.7%), and 2 (2.0%) had one, two, and three follow-up biopsies, respectively. Median time from the initial ASAP diagnosis to the 2nd follow-up biopsy and the last follow-up biopsy were 21.9 months (range 1–129 months) and 27.7 months (range 1–129 months), respectively. Of these patients, 46 (45.1%) were subsequently diagnosed with PCa, including 20 (19.6%) with csPCa. Only 2 (2.0%) patients had GS ≥ 8 disease. Five (4.9%) patients had number of positive cores > 3. Of 35 patients who subsequently underwent RP, seven (20%) had AP after RP and 17 (48.6%) showed GS upgrading. Of these 17 patients, the vast majority (16/17, 94.1%) had GS upgrading from 3 + 3 to 3 + 4. 45.1% of patients with an initial diagnosis of ASAP who had repeat prostate biopsy were subsequently diagnosed with PCa and 19.6% were found to have csPCa. Our findings add further evidence that after a diagnosis of ASAP, a repeat biopsy is warranted and that the repeat biopsy should not be postponed.

A prospective study on inter-operator variability in semi-robotic software-based MRI/TRUS-fusion targeted prostate biopsies

Purpose Magnetic resonance imaging (MRI)/ultrasound-fusion prostate biopsy (FB) comprises multiple steps each of which can cause alterations in targeted biopsy (TB) accuracy leading to false-negative results. The aim was to assess the inter-operator variability of software-based fusion TB by targeting the same MRI-lesions by different urologists. Methods In this prospective study, 142 patients eligible for analysis underwent software-based FB. TB of all lesions (n = 172) were carried out by two different urologists per patient (n = 31 urologists). We analyzed the number of mismatches [overall prostate cancer (PCa), clinically significant PCa (csPCa) and non-significant PCa (nsPCa)] between both performed TB per patient. In addition we evaluated factors contributing to inter-operator variability by uni- and multivariable analyses. Results In 11.6% of all MRI-lesions (10.6% of all patients) there was a mismatch between TB1 and TB2 in terms of overall prostate cancer (PCa detection. Regarding csPCa, patient-based mismatch occurred in 14.8% (n = 21). Overall PCa and csPCa detection rate of TB1 and TB2 did not differ significantly on a per-patient and per-lesion level. Analyses revealed a smaller lesion size as predictive for mismatches (OR 9.19, 95% CI 2.02–41.83, p < 0.001). Conclusion Reproducibility and precision of targeting particularly small lesions is still limited although using software-based FB. Further improvements in image-fusion, segmentation, needle-guidance, and automatization are necessary.

Diagnostic Performance of Ex Vivo Fluorescence Confocal Microscopy in the Assessment of Diagnostic Biopsies of the Prostate

Background: Fluorescence confocal microscopy (FCM) is a novel micro-imaging technique providing optical sections of examined tissue. The method has been well established for the diagnosis of tumors in dermatological specimens. Methods: We compare intraoperative diagnoses of the real-time application of FCM in pre-therapeutic prostate biopsies (35 patients, total number of biopsy specimens: n = 438) with the findings of conventional histology. Results: Prostate carcinoma was reliably diagnosed in all patients. Depending on scan quality and experience of the examiner, smaller lesions of well differentiated carcinoma (ISUP1) could not be consistently differentiated from reactive changes. Furthermore, in some cases there was difficulty to distinguish ISUP grade 2 from ISUP grade 1 tumors. ISUP grades 3–5 were reliably detected in FCM. Conclusion: Despite some limitations, FCM seems to be an effective tool for the timely assessment of prostate biopsies enabling reliable diagnosis of prostate cancer in patients requiring therapy.