Citations

Purpose Radioligand therapy (RLT) with [ 177 Lu] Lu-PSMA-617 has shown significant benefits in metastatic castration-resistant prostate cancer (mCRPC), including prolonged overall survival (OS) and progression-free survival (PFS), as well as improved quality of life. For patients previously treated with [ 177 Lu]Lu-PSMA-617 who subsequently experience disease progression, retreatment with [ 177 Lu] Lu-PSMA-617 has emerged as a potential approach, yet data on the outcomes of RLT retreatment remains limited. This study aimed to assess the efficacy and safety of retreatment RLT in mCRPC. Methods Clinical and imaging data of patients who underwent at least two cycles of [ 177 Lu] Lu-PSMA-617, followed by a second course of treatment with [ 177 Lu] Lu-PSMA-617 between March 2018 and December 2024, were retrospectively reviewed. Endpoints included biochemical response, imaging response, toxicity, PSA-PFS, rPFS, and OS. Radiologic and biochemical responses were evaluated using RECIST 1.1 criteria and PSA levels, respectively. Results A total of 589 patients underwent [ 177 Lu] Lu-PSMA-617 RLT, of whom 20 (3.4%, 20/589) received RLT retreatment. The median age was 71.3 years (range: 63.7–95.2), and the median follow-up was 73.3 months (95% CI, 35.7–77.2 months) after the initial treatment and 21.7 months (95% CI, 16.9–26.0 months) after retreatment. The initial treatment involved a median of 4.5 cycles (range: 2–6). Following the first course of RLT, 15 patients (75%, 15/20) had exhibited a >50% decline in PSA levels, including 9 who achieved a >90% decrease. Additionally, 12 patients (60%, 12/20) had demonstrated a complete or partial imaging response, while 3 (15%, 3/20) patients had stable disease, and 5 (25%, 5/20) patients had progressive radiologic disease. Patients received a median of 4 retreatment cycles (range, 1–6), and the median interval between initial RLT and retreatment RLT was 33.1 months (IQR, 16.2–48.9 months; range, 10.7–58.8 months). A PSA decline of ≥50% was observed in 9 out of 20 patients (45%, 9/20). An imaging response was observed in 10 patients (50%, 10/20), with 4 achieving a complete response and 6 achieving a partial response, whereas 10 patients (50%, 10/20) showed progression. The median PSA-PFS was 9.8 months (95% CI, 4.1–17.9 months), and the median rPFS was 8.2 months (95% CI, 3.3–17.9 months). Eight deaths occurred during follow-up. The estimated Kaplan Meier OS rates were 83.6% at 12 months (95% CI, 57.3%–94.4%) and 53.7% at 24 months (95% CI, 28.3%–73.7%). Grade 2 renal toxicity was observed in one patient (5%, 1/20). Three patients (15%, 3/20) developed Grade 2 anemia. Notably, no grade 3 or grade 4 adverse events were observed during either treatment period. Conclusion In our cohort, retreatment with [¹ 77 Lu]Lu-PSMA-617 RLT was associated with biochemical and imaging responses in some patients, with observed survival outcomes that support further investigation. However, given the retrospective single-arm design and small sample size, large prospective studies are needed to more rigorously evaluate the safety, efficacy, and optimal patient selection for retreatment RLT.

Abstract Glioblastoma (GBM), isocitrate dehydrogenase wild-type (IDHwt) and central nervous system diffuse large B-cell lymphoma (CNS-DLBCL) are aggressive brain tumors with overlapping MRI features, yet distinct treatment approaches. Noninvasive tools are needed to aid in differential diagnosis. Deep learning on T1 postcontrast and T2-weighted MRI sequences were used to differentiate GBM and CNS-DLBCL. A three-stage temporal study design was utilized. Model development was performed on 146 patients with CNS-DLBCL and 146 age-matched, sex-matched, and MRI year–matched patients with GBM diagnosed at Mayo Clinic between 1998 and 2019. Models were tested on independent temporal test cohorts. Initial testing included 240 independent GBM diagnosed at Mayo Clinic between 1998 and 2019. The prospective test cohort included 37 patients with CNS-DLBCL and 256 patients with GBM diagnosed at Mayo Clinic after January 1, 2020, and 36 patients with CNS-DLBCL diagnosed at an external institution. Of the patients diagnosed at Mayo Clinic, 47% had MRIs generated from non-Mayo institutions. Two different model approaches were compared: (i) ensemble approach using area under the receiver operating characteristic curve (AUC) and cross-validation for model selection and (ii) loss approach minimizing cross-entropy loss and cross-validation to evaluate prediction performance. The AUCs on the prospective test cohort were 0.84 [95% confidence interval (CI), 0.78–0.90] and 0.83 (95% CI, 0.77–0.88) for the ensemble and loss approaches, respectively. Stability of ensemble prediction improved with the increasing number of models. Stratified AUC analysis demonstrated consistent performance across sex and age. We utilized a robust temporal study design and applied 2 different analytic approaches to develop a classification model. The findings confirm the feasibility of using MRI-based deep learning models to differentiate GBM from CNS-DLBCL. Significance: GBM, IDHwt and CNS-DLBCL are aggressive brain tumors with overlapping MRI features, yet distinct treatment approaches. Noninvasive tools are needed to aid in differential diagnosis. We developed MRI-based deep learning models to differentiate GBM, IDHwt from CNS-DLBCL using a rigorous three-stage temporal design that included prospective validation. The model AUC on a prospective cohort was 0.84.

ABSTRACT Circulating tumor DNA (ctDNA) as a liquid biopsy is an emerging tool used for diagnosis and disease monitoring, yet not all tumors shed DNA equally, thus currently limiting the clinical applicability of this tool. Our institution has developed and validated a droplet digital Polymerase Chain Reaction (ddPCR) ctDNA test for BRAF V600 mutation. We collected data on all BRAF mutated melanoma patients tested with the ddPCR BRAF assay at our institution over the years 2018–2024 and correlated the test results with clinical and radiological baseline parameters, with a focus on Positron Emission Tomography‐Computed Tomography (PET‐CT) data, incorporating both location of lesions and volume of disease. Linear and logistic regression modelling was used to assess possible associations between the baseline parameters and the ctDNA results. We identified 71 BRAF mutated melanoma patients, of which 65 had active disease on imaging studies and 43 had PET‐CT data available for analysis. Multivariate modelling has shown that patients were more likely to have a positive ctDNA result if they had elevated serum lactate dehydrogenase (LDH) (OR = 9.9), had a higher number of lesions on imaging studies (> 10 lesions: OR = 11.3), and if they were younger than 65 (OR = 11.8), with an AUC of 0.84. The PET‐CT calculated total metabolic tumor volume correlated with the quantitative ctDNA result with a Pearson r factor of 0.49. In conclusion, ctDNA ddPCR BRAF testing is easily obtained and might be especially beneficial for patients diagnosed with a clinically aggressive disease, based on elevated LDH, higher PET‐CT metabolic tumor volume, and younger age.

Background and Objective: Radiographic progression in prostate cancer (PCa) can occur even when prostate-specific antigen (PSA) levels are undetectable. We aimed to determine the frequency and characteristics of radiographic disease progression (rDP) on PSMA PET/CT in patients with undetectable PSA, referred to as PSA zero rDP. Methods: We analyzed the Mayo Clinic PSMA PET Prostate Cancer Registry to identify patients with rDP on PSMA PET/CT despite undetectable PSA levels. Disease progression was confirmed via biopsy or treatment response. The cohort included patients with non-metastatic and metastatic hormone-sensitive disease, as well as those with castration-resistant prostate cancer at the time of imaging. Overall survival (OS) was estimated using the Kaplan–Meier method. Group comparisons were performed with the log-rank test. Univariate Cox regression was used to identify factors associated with poor OS. Key findings and Limitations: Among 2141 patients imaged between 2021 and 2023, 257 (12%) had PSA zero rDP. Sixty-one percent had initially localized disease; 39% had de novo metastatic disease. Median (IQR) time from diagnosis to PSA zero rDP was 51.9 (18.4–115.5) months. A total of 184 patients (72%) progressed to castration-resistant PCa. Sites of rDP included bone (57%), visceral (15%), lymph node (18%), and local recurrence (10%). During median follow-up of 8.1 (3.5–11.9) months, 5% of patients died. Only visceral metastases were significantly associated with poorer OS (p < 0.0001). Conclusions and Clinical Implications: Prostate cancer patients frequently develop metastatic disease with undetectable PSA values. Our findings suggest the use of periodic advanced imaging techniques, irrespective of PSA value, for more prompt detection and early management of disease progress.

The translation of AI-generated brain metastases (BM) segmentation into clinical practice relies heavily on diverse, high-quality annotated medical imaging datasets. The BraTS-METS 2023 challenge has gained momentum for testing and benchmarking algorithms using rigorously annotated internationally compiled real-world datasets. This study presents the results of the segmentation challenge and characterizes the challenging cases that impacted the performance of the winning algorithms. Untreated brain metastases on standard anatomic MRI sequences (T1, T2, FLAIR, T1PG) from eight contributed international datasets were annotated in stepwise method: published UNET algorithms, student, neuroradiologist, final approver neuroradiologist. Segmentations were ranked based on lesion-wise Dice and Hausdorff distance (HD95) scores. False positives (FP) and false negatives (FN) were rigorously penalized, receiving a score of 0 for Dice and a fixed penalty of 374 for HD95. The mean scores for the teams were calculated. Eight datasets comprising 1303 studies were annotated, with 402 studies (3076 lesions) released on Synapse as publicly available datasets to challenge competitors. Additionally, 31 studies (139 lesions) were held out for validation, and 59 studies (218 lesions) were used for testing. Segmentation accuracy was measured as rank across subjects, with the winning team achieving a LesionWise mean score of 7.9. The Dice score for the winning team was 0.65 ± 0.25. Common errors among the leading teams included false negatives for small lesions and misregistration of masks in space. The Dice scores and lesion detection rates of all algorithms diminished with decreasing tumor size, particularly for tumors smaller than 100 mm3. In conclusion, algorithms for BM segmentation require further refinement to balance high sensitivity in lesion detection with the minimization of false positives and negatives. The BraTS-METS 2023 challenge successfully curated well- annotated, diverse datasets and identified common errors, facilitating the translation of BM segmentation across varied environments and providing the tools for future development of personalized volumetric reports to patients undergoing BM treatment.

e15046 Background: BRAF V600 is the most common driver mutation in melanoma, seen in ≈50% of patients. Recently, the use of liquid biopsy assays that are based on circulating tumor DNA (ctDNA) has become widespread for diagnosis and residual disease monitoring. We have previously reported on our droplet digital PCR (ddPCR) test as a liquid biopsy for BRAF mutant (BRAFm) melanoma. Here we report the correlation between clinical and radiological parameters and ddPCR test results. Methods: Using the department of laboratory medicine archives, we have identified all biopsy proven BRAFm melanoma patients that were tested with the BRAF V600 ddPCR assay at our institution over the years 2018-2024. All patients had blood samples collected, and cell free DNA (cfDNA) was extracted from plasma using the Qiagen QIAamp Circulating Nucleic Acid Kit. cfDNA was evaluated with a BioRad QX200 droplet digital PCR platform assessing for BRAF V600E/K mutations. Clinical and demographic data were collected from medical records. Radiometabolic features including volumetric data and FDG PET SUV values were collected and analyzed using the MIM software (MIM Software Inc., Beachwood, OH, USA). Univariate and multivariate analyses were used to assess correlations between baseline variables and test results. Results: 71 patients were identified. The median age was 59 (26-85), 54% were males. 46 patients (65%) had a positive test result which was based on a cut-off of ≥4 mutant copies per ml. 65 patients (92%) had active disease on imaging studies, of which those with a positive test result were younger (p = 0.047), had a higher proportion of osseous metastases (p = 0.02) and had a higher number of lesions (p = 0.006). There was no difference in the median overall survival between the groups (11 months vs. 33 months for positive and negative result respectively, log-rank test; p = 0.15). Total tumor volume on PET-CT (in ml) was predictive of the quantitative ddPCR test result (in copies per ml), with every increase in 10 ml of tumor leading to an expected increase of 370 copies per ml (p < 0.001). Multivariate models have failed to show a correlation between baseline parameters and the quantitative ddPCR test result. When performing the analyses using logistic regression models with the binary ddPCR test result as the outcome, osseous metastases were predictive in a univariate model (OR = 4.3, 95% CI 1.4-15.4). The best fit multivariate model showed that higher burden of disease with ≥6 lesions (OR = 6.57, 95% CI 1.24-44.3) and elevated LDH (OR = 8.8, 95% CI 1.83-56.91) were associated with an increased risk for a positive result, while younger age was associated with decreased risk for a positive result (OR = 0.92, 95% CI 0.86-0.97). Conclusions: The results of ddPCR testing for BRAF mutation are correlated with the radiological and volumetric extent of the disease. This test could be useful in patients presenting with high-burden disease when a fast turnaround time is needed.

114 Background: Oncologists have traditionally relied upon serial assessments of PSA and periodic imaging (i.e., every 2 to 3 cycles) to monitor treatment response. Early identification of disease progression allows for faster transitions to alternative therapies that might be more effective. In this study, we evaluated the utility of post-treatment SPECT/CT for earlier detection of treatment failure in patients receiving Lutetium-177–PSMA-617 (LuPSMA). Methods: We queried an IRB-approved registry including all patients starting LuPSMA at Mayo Clinic, MN, in the interval of March 2022 to March 2023. Those receiving fewer than 2 cycles of treatment and missing SPECT/CT data were excluded. We reviewed clinical notes to document the date and reason(s) for treatment discontinuation. The assembled cohort consists of patients stopping LuPSMA for progressive disease ([PD], categorized as biochemical, radiographic, or both). Results of the prior SPECT/CT imaging reports were reviewed. At our institution SPECT/CT images are typically acquired on the day after each infusion. Metrics collected from the radiology reports included: new foci of PSMA localization since the prior PSMA PET or SPECT/CT, suspicion for new non-PSMA avid disease, and a visual determination of PSMA-avid tumor volume. Results: A total of 256 patients received an initial cycle of LuPSMA. SPECT/CT imaging data for at least 2 consecutive treatment cycles was available for 68 patients who developed PD after a median (IQR) of 3 (2-4) cycles. There were 55 patients (81%) with both biochemical and radiographic evidence of PD, 7 (10%) with biochemical only progression, and 6 (9%) with radiographic only progression. The most used imaging modality for response assessment was PSMA PET/CT (n=46, 67%), followed by conventional imaging (n=9, 13%), and choline PET/CT (n=5, 7%). The median time between most recent post-therapy SPECT/CT and determination of PD was 39 days. The prior SPECT/CT had detected new foci of PSMA avid disease in 12 (18%) cases and new non-PSMA avid disease in 5 (7%) cases. Two patients had both new PSMA-avid and non-PSMA avid lesions. The most common sites of new PSMA-avid disease on SPECT/CT were bone (n=10) and liver (n=2). A total of 20 (29%) patients had visually increased tumor volume on SPECT/CT, 9 (45%) of which had no new lesions. In total, 24 of 68 patients (35%) had early detection of treatment futility detected by SPECT/CT. At a median (IQR) follow-up time of 8 (6.4-12) mo, the median OS [95% CI] was similar for patients with PD detected on SPECT/CT (7.8 [6.6 – 9] mo) compared to those without clear evidence of PD on SPECT/CT(8.4 [6.4-10.3] mo, p=0.885). Conclusions: Post-treatment SPECT/CT imaging may complement traditional forms of response assessment and provide an earlier warning of treatment failure through identification of new PSMA-avid lesions, non-PSMA avid disease, and increased tumor volume.

175 Background: 177 Lu-PSMA-617, a PSMA-directed radionuclide therapy, is an emerging treatment for metastatic castration-resistant prostate cancer (mCRPC). This retrospective study evaluates clinical outcomes of 177 Lu-PSMA-617 in low volume (LV) and high volume (HV) mCRPC, classified using the CHAARTED criteria. Methods: We conducted a retrospective review of the Mayo Clinic Prostate Cancer Registry, including mCRPC patients treated with 177 Lu-PSMA-617 at Mayo Clinic Rochester, MN, from August 2017 to August 2024. Patients were categorized based on CHAARTED criteria, which define high-volume disease by the presence of visceral metastases and/or four or more bone metastases, with at least one outside the vertebral column and pelvis. PSA-response rate (PSA-RR), overall survival (OS), PSA-progression-free survival (PSA-PFS), and radiographic progression-free survival (rPFS) were analyzed using Kaplan-Meier curves with log-rank tests, as well as uni- and multivariate Cox regression. Results: Of the 264 patients treated with 177 Lu-PSMA-617, 83 were classified as LV (median follow-up: 17.1 months) and 177 as HV (median follow-up: 13.7 months). LV patients had lower serum PSA levels (P < 0.0001), higher hemoglobin levels (P < 0.0001), and a longer time from diagnosis to treatment initiation (10.3 years vs. 7.2 years, P = 0.04). In the HV group, 92% had bone metastases, 60% had lymph node involvement, 54% had both, and 35% had visceral metastases. In the LV group, 64% had bone metastases, 84% had lymph node involvement, 25% had both, and none had visceral metastases. Median OS was 14.4 months for HV patients and 17.6 months for LV patients (P < 0.0001). Median PSA-PFS was 11.4 months for HV patients and 13 months for LV patients. Median rPFS was 9.1 months for HV patients and 10.6 months for LV patients. PSA-RR significantly differed between groups (P < 0.0001). In the HV group, 20% achieved CR (PSA < 0.2 ng/mL), 29% had PR (PSA decreased by ≥50%), 16% experienced SD, and 35% had PD (PSA increased ≥25% from nadir). In contrast, the LV group had 51% achieving CR, 10% achieving PR, 18% with SD, and 11% with PD. Conclusions: Lower disease volume in mCRPC is strongly associated with improved outcomes following 177 Lu-PSMA-617 therapy. These findings suggest that tumor burden is a key predictor of response to PSMA radioligand therapy.

Abstract Clinical management of IDH wildtype glioblastoma (GBM) and tumefactive multiple sclerosis (tMS) is drastically different. GBM requires maximal safe resection followed by chemoradiation, while tMS outcome is worsened by surgery and radiotherapy. Noninvasive methods are needed to help with accurate diagnosis of tumor and non-tumor etiologies. To develop an MRI-based classification model, tMS subjects diagnosed prior to January 1, 2020, were matched to tMS by age at diagnosis, sex, index MRI date, and 2D/3D acquisition. Inclusion criteria included one cm minimal lesion size and pre-operative post-contrast T1 and T2 images available for analysis. A 3D-DenseNet121 was used to develop a classification model using prespecified parameters: 650 epochs, batch size 16, learning rate 10-3, cross-entropy loss, and AdamW optimizer. The stopping rule was defined as three sequential differences in epoch cross-entropy loss &lt;0.02. Models were developed using both T1gd and T2, as well as from only T1gd and only T2. Training included 220 subjects (110 GBM, 110 tMS). A 2-stage validation design was used, which included both retrospective and prospective cohorts. Stage 1 consisted of 272 retrospective GBM (diagnosed prior to January 1, 2020). Stage 2 consisted of 69 and 34 prospective (diagnosed after January 1, 2020) GBM and tMS, respectively. External validation on the 272 retrospective GBM demonstrated accuracy of 91%, 84%, and 78% for T1gd+T2, T1gd only, and T2 only, respectively. External validation on the 69 prospective GBM demonstrated an accuracy of 87%, 64%, and 67% for T1gd+T2, T1gd only, and T2 only, respectively. The 34 prospective tMS demonstrated accuracy of 76%, 76%, and 82% for T1gd+T2, T1gd only, and T2 only, respectively. This shows the feasibility of deep learning to aid in differential diagnosis of brain lesions. Future work entails the integration of germline variants into the classification model, including variants associated with the risk of glioma or MS.