Conceptual framework and documentation standards of cystoscopic media content for artificial intelligence.

BACKGROUND: The clinical documentation of cystoscopy includes visual and textual materials. However, the secondary use of visual cystoscopic data for educational and research purposes remains limited due to inefficient data management in routine clinical practice. METHODS: A conceptual framework was designed to document cystoscopy in a standardized manner with three major sections: data management, annotation management, and utilization management. A Swiss-cheese model was proposed for quality control and root cause analyses. We defined the infrastructure required to implement the framework with respect to FAIR (findable, accessible, interoperable, reusable) principles. We applied two scenarios exemplifying data sharing for research and educational projects to ensure compliance with FAIR principles. RESULTS: The framework was successfully implemented while following FAIR principles. The cystoscopy atlas produced from the framework could be presented in an educational web portal; a total of 68 full-length qualitative videos and corresponding annotation data were sharable for artificial intelligence projects covering frame classification and segmentation problems at case, lesion, and frame levels. CONCLUSION: Our study shows that the proposed framework facilitates the storage of visual documentation in a standardized manner and enables FAIR data for education and artificial intelligence research.

An Efficient Framework for Video Documentation of Bladder Lesions for Cystoscopy: A Proof-of-Concept Study.

Processing full-length cystoscopy videos is challenging for documentation and research purposes. We therefore designed a surgeon-guided framework to extract short video clips with bladder lesions for more efficient content navigation and extraction. Screenshots of bladder lesions were captured during transurethral resection of bladder tumor, then manually labeled according to case identification, date, lesion location, imaging modality, and pathology. The framework used the screenshot to search for and extract a corresponding 10-seconds video clip. Each video clip included a one-second space holder with a QR barcode informing the video content. The success of the framework was measured by the secondary use of these short clips and the reduction of storage volume required for video materials. From 86 cases, the framework successfully generated 249 video clips from 230 screenshots, with 14 erroneous video clips from 8 screenshots excluded. The HIPPA-compliant barcodes provided information of video contents with a 100% data completeness. A web-based educational gallery was curated with various diagnostic categories and annotated frame sequences. Compared with the unedited videos, the informative short video clips reduced the storage volume by 99.5%. In conclusion, our framework expedites the generation of visual contents with surgeon's instruction for cystoscopy and potential incorporation of video data towards applications including clinical documentation, education, and research.

Specific spatial distribution patterns of tumor foci are associated with a low risk of biochemical recurrence in pT2pN0R0 prostate cancer.

BACKGROUND: The previous attempts for pT2 substaging of prostate cancer (PCa) were insufficient in providing prognostic subgroups and the search for new prognostic parameters to subcategorize pT2 PCa is, therefore, needed. Therefore, the current study investigated the association between tumor distribution patterns and the biochemical recurrence (BCR)-free survival rate in pT2pN0R0 PCa. METHODS: Following radical prostatectomy, the anatomical distribution of PCa in 743 men with pT1-pT3pN0 disease was analyzed to determine 20 types of PCa distribution patterns. Then, 245 men with pT2pN0R0 PCa was considered for prognostic evaluation with a mean follow-up period of 60 months. The spatial distribution patterns of PCa were evaluated using a cMDX©-based map model of the prostate. An analysis including 552,049 comparison operations was performed to assist in the evaluation of the similarity levels of the distribution patterns. A k-mean cluster analysis was applied to determine groups with similar distribution patterns. A decision-tree analysis was performed to divide these groups according to frequency of BCR. The BCR-free survival rate was analyzed using Kaplan-Meier curves. Predictors of progression were investigated using a Cox proportional hazards model. RESULTS: BCR occurred in 8.2% of the 245 men with pT2pN0R0 PCa. The median time of recurrence was 60 months (interquartile range [IQR]: 42-77). In univariate and multivariate analyses, the prostate volume and the distribution patterns were independent predictors for BCR, whereas the sub-staging of pT2 tumors, Gleason grading, prostate-specific antigen (PSA) level, and relative tumor volume were not. In the patients with pT2pN0R0 disease, PCa distribution patterns with the apical involvement were significantly associated with the risk of BCR (P = 0.001). CONCLUSION: The spread tumor patterns with the apical involvement are associated with a high-risk of BCR in the pT2 tumor stage. The vertical tumor spread could be considered in developing improved prognostic pT2 sub-categories.

Blind Biobanking of the Prostatectomy Specimen: Critical Evaluation of the Existing Techniques and Development of the New 4-Level Tissue Extraction Model With High Sampling Efficacy.

BACKGROUND: Fresh tissue is mandatory to perform high-quality translation studies. Several models for tissue extraction from prostatectomy specimens without guidance by frozen sections are already introduced. However, little is known about the sampling efficacy of these models, which should provide representative tissue in adequate volumes, account for multifocality and heterogeneity of tumor, not violate the routine final pathological examination, and perform quickly without frozen section-based histological control. The aim of the study was to evaluate the sampling efficacy of the existing tissue extraction models without guidance by frozen sections ("blind") and to develop an optimized model for tissue extraction. METHODS: Five hundred thirty-three electronic maps of the tumor distribution in prostates from a single-center cohort of the patients subjected to radical prostatectomy were used for analysis. Six available models were evaluated in silico for their sampling efficacy. Additionally, a novel model achieving the best sampling efficacy was developed. RESULTS: The available models showed high efficacies for sampling "any part" from the tumor (up to 100%), but were uniformly low in efficacy to sample all tumor foci from the specimens (with the best technique sampling only 51.6% of the all tumor foci). The novel 4-level extraction model achieved a sampling efficacy of 93.1% for all tumor foci. CONCLUSIONS: The existing "blind" tissue extraction models from prostatectomy specimens without frozen sections control are suitable to target tumor tissues but these tissues do not represent the whole tumor. The novel 4-level model provides the highest sampling efficacy and a promising potential for integration into routine. Prostate 77: 396-405, 2017. © 2016 Wiley Periodicals, Inc.

Postoperative Nomogram for Relapse-Free Survival in Patients with High Grade Upper Tract Urothelial Carcinoma.

PURPOSE: We developed a prognostic nomogram for patients with high grade urothelial carcinoma of the upper urinary tract after extirpative surgery. MATERIALS AND METHODS: Clinical data were available for 2,926 patients diagnosed with high grade urothelial carcinoma of the upper urinary tract who underwent extirpative surgery. Cox proportional hazard regression models identified independent prognosticators of relapse in the development cohort (838). A backward step-down selection process was applied to achieve the most informative nomogram with the least number of variables. The L2-regularized logistic regression was applied to generate the novel nomogram. Harrell's concordance indices were calculated to estimate the discriminative accuracy of the model. Internal validation processes were performed using bootstrapping, random sampling, tenfold cross-validation, LOOCV, Brier score, information score and F1 score. External validation was performed on an external cohort (2,088). Decision tree analysis was used to develop a risk classification model. Kaplan-Meier curves were applied to estimate the relapse rate for each category. RESULTS: Overall 35.3% and 30.7% of patients experienced relapse in the development and external validation cohort. The final nomogram included age, pT stage, pN stage and architecture. It achieved a discriminative accuracy of 0.71 and 0.76, and the AUC was 0.78 and 0.77 in the development and external validation cohort, respectively. Rigorous testing showed constant results. The 5-year relapse-free survival rates were 88.6%, 68.1%, 40.2% and 12.5% for the patients with low risk, intermediate risk, high risk and very high risk disease, respectively. CONCLUSIONS: The current nomogram, consisting of only 4 variables, shows high prognostic accuracy and risk stratification for patients with high grade urothelial carcinoma of the upper urinary tract following extirpative surgery, thereby adding meaningful information for clinical decision making.

Dynamic changes of alkaline phosphatase are strongly associated with PSA-decline and predict best clinical benefit earlier than PSA-changes under therapy with abiraterone acetate in bone metastatic castration resistant prostate cancer.

BACKGROUND: Significant progress in treatment of metastatic castration resistant prostate cancer (mCRPC) has been made. Biomarkers to tailor therapy are scarce. To facilitate decision-making we evaluated dynamic changes of alkaline phosphatase (ALP), lactate dehydrogenase (LDH) and prostate specific antigen (PSA) under therapy with Abiraterone. METHODS: Men with bone mCRPC (bmCRPC) on Abiraterone 12/2009-01/2014 were analyzed. Dynamic ALP-, LDH- and PSA-changes were analyzed as predictors of best clinical benefit and overall survival (OS) with logistic-regression, Cox-regression and Kaplan-Meier-analysis. RESULTS: Thirty-nine pre- and 45 post-chemotherapy patients with a median follow up of 14.0 months were analyzed. ALP-Bouncing can be observed very early during therapy with Abiraterone. ALP-Bouncing is defined as rapidly rising ALP-levels independent of baseline ALP during the first 2-4 weeks of Abiraterone-therapy with subsequent equally marked decline to pretreatment levels or better within 8 weeks of therapy, preceding potentially delayed PSA-decline. In univariate analysis failure of PSA-reduction ≥ 50% and failure of ALP-Bouncing were the strongest predictors of progressive disease (p = 0.003 and 0.021). Rising ALP at 12 weeks, no PSA-reduction ≥ 50% and no ALP-Bouncing were strongest predictors of poor OS, (all p < 0.001). Kaplan-Meier-analysis showed worse OS for rising ALP at 12 weeks, no PSA-reduction ≥ 50% and no ALP-Bouncing (p < 0.001). In subgroup-analysis of oligosymptomatic patients all parameters remained significant predictors of poor OS, with no PSA-reduction ≥ 50% and rising ALP at 12 weeks being the strongest (p < 0.001). In multivariate analysis PSA-reduction ≥ 50% remained an independent predictor of OS for the whole cohort and for the oligosymptomatic subgroup (both p = 0.014). No patient with ALP-Bouncing had PD for best clinical benefit. Patients with rising ALP at 12 weeks had no further benefit of Abiraterone. CONCLUSIONS: Dynamic changes of ALP, LDH and PSA during Abiraterone-therapy are associated with best clinical benefit and OS in bmCRPC. ALP-Bouncing occurring earlier than PSA-changes as well as prior to equivocal imaging results and rising ALP at 12 weeks under Abiraterone may help to decide whether to discontinue Abiraterone. An external validation of these findings on a prospective cohort is planned.

Analysis of topographical distribution of prostate cancer and related pathological findings in prostatectomy specimens using cMDX document architecture.

INTRODUCTION: Understanding the topographical distribution of prostate cancer (PCa) foci is necessary to optimize the biopsy strategy. This study was done to develop a technical approach that facilitates the analysis of the topographical distribution of PCa foci and related pathological findings (i.e., Gleason score and foci dimensions) in prostatectomy specimens. MATERIAL & METHODS: The topographical distribution of PCa foci and related pathologic evaluations were documented using the cMDX documentation system. The project was performed in three steps. First, we analyzed the document architecture of cMDX, including textual and graphical information. Second, we developed a data model supporting the topographic analysis of PCa foci and related pathologic parameters. Finally, we retrospectively evaluated the analysis model in 168 consecutive prostatectomy specimens of men diagnosed with PCa who underwent total prostate removal. The distribution of PCa foci were analyzed and visualized in a heat map. The color depth of the heat map was reduced to 6 colors representing the PCa foci frequencies, using an image posterization effect. We randomly defined 9 regions in which the frequency of PCa foci and related pathologic findings were estimated. RESULTS: Evaluation of the spatial distribution of tumor foci according to Gleason score was enabled by using a filter function for the score, as defined by the user. PCa foci with Gleason score (Gls) 6 were identified in 67.3% of the patients, of which 55 (48.2%) also had PCa foci with Gls between 7 and 10. Of 1173 PCa foci, 557 had Gls 6, whereas 616 PCa foci had Gls>6. PCa foci with Gls 6 were mostly concentrated in the posterior part of the peripheral zone of the prostate, whereas PCa foci with Gls>6 extended toward the basal and anterior parts of the prostate. The mean size of PCa foci with Gls 6 was significantly lower than that of PCa with Gls>6 (P<0.0001). CONCLUSION: The cMDX-based technical approach facilitates analysis of the topographical distribution of PCa foci and related pathologic findings in prostatectomy specimens.

Prostate cancers detected on repeat prostate biopsies show spatial distributions that differ from those detected on the initial biopsies.

OBJECTIVE: To evaluate the spatial distribution of prostate cancer detected at a single positive biopsy (PBx) and a repeat PBx (rPBx). PATIENTS AND METHODS: We evaluated 533 consecutive men diagnosed with prostate cancer who underwent radical prostatectomy using a clinical map document based on XML (cMDX©)-based map model of the prostate. We determined the number of cancer foci, relative tumour volume, Gleason score, zone of origin, localisation, and pathological stage after stratification according to the number of PBx sessions (PBx vs rPBx). The distribution of 3966 prostate cancer foci was analysed and visualised on heat maps. The colour gradient of the heat map was reduced to six colours representing the frequency classification of prostate cancer using an image posterisation effect. Additionally, the spatial distribution of organ-confined prostate cancer between PBx and rPBx was evaluated. RESULTS: Prostate cancer diagnosed on PBx was mostly localised to the apical portion and the peripheral zone of the prostate. Prostate cancer diagnosed on rPBx was more frequently found in the anterior portion and the base of the prostate. Organ-confined prostate cancer foci were mostly localised in the dorsolateral zone of the prostate in men at PBx, whereas men at rPBx had more prostate cancer foci in the anterior portion. CONCLUSIONS: The spatial distribution of prostate cancer with rPBx differs significantly from the spatial distribution of prostate cancer with PBx. The whole anterior portion of the prostate should be considered by rPBx.

Treatment of Incidental Prostate Cancer by Active Surveillance: Results of the HAROW Study.

OBJECTIVE: To report on a cohort of patients with incidental prostate cancer (IPC) that was treated by an active surveillance (AS) protocol in the HAROW study. MATERIALS AND METHODS: The HAROW study is an observational study on the management of localized prostate cancer in Germany. Treating urologists were reporting clinical parameters, information on therapy and clinical course of disease at 6-month intervals. RESULTS: In total, 3,169 patients were enrolled. In 224 patients were found an IPC and 104 (46%) of them were put on an AS protocol. The mean follow-up was 26.5 months. Tumor progression was noted in 16 patients. In 11 patients, AS was replaced by a definite intervention. In univariate and multivariate analyses, only PSA density correlated with progression. CONCLUSION: This is the first prospective description of an IPC patient cohort on AS as part of an outcomes research study. AS was selected as a therapeutic strategy in nearly half of the patients (46%). Only a minor proportion (16%) displayed progression. Of the clinical parameters, only PSA density correlated with progression.

Deep learning identifies histopathologic changes in bladder cancers associated with smoke exposure status.

Smoke exposure is associated with bladder cancer (BC). However, little is known about whether the histologic changes of BC can predict the status of smoke exposure. Given this knowledge gap, the current study investigated the potential association between histology images and smoke exposure status. A total of 483 whole-slide histology images of 285 unique cases of BC were available from multiple centers for BC diagnosis. A deep learning model was developed to predict the smoke exposure status and externally validated on BC cases. The development set consisted of 66 cases from two centers. The external validation consisted of 94 cases from remaining centers for patients who either never smoked cigarettes or were active smokers at the time of diagnosis. The threshold for binary categorization was fixed to the median confidence score (65) of the development set. On external validation, AUC was used to assess the randomness of predicted smoke status; we utilized latent feature presentation to determine common histologic patterns for smoke exposure status and mixed effect logistic regression models determined the parameter independence from BC grade, gender, time to diagnosis, and age at diagnosis. We used 2,000-times bootstrap resampling to estimate the 95% Confidence Interval (CI) on the external validation set. The results showed an AUC of 0.67 (95% CI: 0.58-0.76), indicating non-randomness of model classification, with a specificity of 51.2% and sensitivity of 82.2%. Multivariate analyses revealed that our model provided an independent predictor for smoke exposure status derived from histology images, with an odds ratio of 1.710 (95% CI: 1.148-2.54). Common histologic patterns of BC were found in active or never smokers. In conclusion, deep learning reveals histopathologic features of BC that are predictive of smoke exposure and, therefore, may provide valuable information regarding smoke exposure status.

Electronic Documentation of Intraoperative Observation of Cystoscopic Procedures Using the cMDX Information System.

PURPOSE: Accurate documentation of lesions during transurethral resection of bladder tumors (TURBT) is essential for precise diagnosis, treatment planning, and follow-up care. However, optimizing schematic documentation techniques for bladder lesions has received limited attention. MATERIALS AND METHODS: This prospective observational study used a cMDX-based documentation system that facilitates graphical representation, a lesion-specific questionnaire, and heatmap analysis with a posterization effect. We designed a graphical scheme for bladder covering bladder landmarks to visualize anatomic features and to document the lesion location. The lesion-specific questionnaire was integrated for comprehensive lesion characterization. Finally, spatial analyses were applied to investigate the anatomic distribution patterns of bladder lesions. RESULTS: A total of 97 TURBT cases conducted between 2021 and 2023 were included, identifying 176 lesions. The lesions were distributed in different bladder areas with varying frequencies. The distribution pattern, sorted by frequency, was observed in the following areas: posterior, trigone, lateral right and anterior, and lateral left and dome. Suspicious levels were assigned to the lesions, mostly categorized either as indeterminate or moderate. Lesion size analysis revealed that most lesions fell between 5 and 29 mm. CONCLUSION: The study highlights the potential of schematic documentation techniques for informed decision making, quality assessment, primary research, and secondary data utilization of intraoperative data in the context of TURBT. Integrating cMDX and heatmap analysis provides valuable insights into lesion distribution and characteristics.

Critical evaluation of artificial intelligence as a digital twin of pathologists for prostate cancer pathology.

Prostate cancer pathology plays a crucial role in clinical management but is time-consuming. Artificial intelligence (AI) shows promise in detecting prostate cancer and grading patterns. We tested an AI-based digital twin of a pathologist, vPatho, on 2603 histological images of prostate tissue stained with hematoxylin and eosin. We analyzed various factors influencing tumor grade discordance between the vPatho system and six human pathologists. Our results demonstrated that vPatho achieved comparable performance in prostate cancer detection and tumor volume estimation, as reported in the literature. The concordance levels between vPatho and human pathologists were examined. Notably, moderate to substantial agreement was observed in identifying complementary histological features such as ductal, cribriform, nerve, blood vessel, and lymphocyte infiltration. However, concordance in tumor grading decreased when applied to prostatectomy specimens (κ = 0.44) compared to biopsy cores (κ = 0.70). Adjusting the decision threshold for the secondary Gleason pattern from 5 to 10% improved the concordance level between pathologists and vPatho for tumor grading on prostatectomy specimens (κ from 0.44 to 0.64). Potential causes of grade discordance included the vertical extent of tumors toward the prostate boundary and the proportions of slides with prostate cancer. Gleason pattern 4 was particularly associated with this population. Notably, the grade according to vPatho was not specific to any of the six pathologists involved in routine clinical grading. In conclusion, our study highlights the potential utility of AI in developing a digital twin for a pathologist. This approach can help uncover limitations in AI adoption and the practical application of the current grading system for prostate cancer pathology.

High-grade prostatic intraepithelial neoplasia (HGPIN) and topographical distribution in 1,374 prostatectomy specimens: existence of HGPIN near prostate cancer.

PURPOSE: High-grade prostatic intraepithelial neoplasia (HGPIN) is believed to be a precursor of prostate cancer (PCa). This study evaluated whether HGPIN was located close to PCa in whole radical prostatectomy specimens (RPSs). MATERIALS AND METHODS: We evaluated 1,374 prostate specimens from 1999 to 2010 using a cMDX-based map model of the prostate. The distribution of 10,439 PCa foci was analyzed and visualized on a heat map. The color gradient of the heat map was reduced to six colors representing the frequency classification of the relative frequency of PCa using an image posterization effect. We defined 22 regions in the prostate according to the frequency of PCa occurrence. Seven hundred ninety RPSs containing 6,374 PCa foci and 4,502 HGPIN foci were evaluated. The topographical association between PCa and HGPIN in the RPSs was analyzed by estimating the frequencies of PCa and HGPIN in 22 regions. A logistic regression analysis was performed to assess the odds ratios of HGPIN for the presence of PCa in 22 regions. RESULTS: Fifty-eight percent of PCa specimens included HGPIN and had significantly more favorable Gleason scores, lower PSA levels and smaller relative tumor volumes than isolated PCa specimens. HGPIN (68%) and PCa (69%) were predominantly localized to the apical half of the prostate. HGPIN was mainly concentrated in the peripheral zone medial to regions with high PCa frequencies. Upon logistic regression analysis, HGPIN was a significant predictor of PCa co-existence in 11 regions. CONCLUSIONS: HGPIN was located adjacent to PCa in whole RPSs. PCa concomitant with HGPIN had more favorable pathologic features than isolated PCa.

Linkage of data from diverse data sources (LDS): a data combination model provides clinical data of corresponding specimens in biobanking information system.

To provide sufficient clinical data for corresponding specimens from diverse databases established before the implementation of biobanks for research purposes with respect to data privacy regulations. For this purpose, we developed a data model called "linkage of data from diverse data sources (LDS)". The data model was developed to merge clinical data from an existing local database with biospecimen repository data in our serum bank for uro-oncology. This concept combines two data models based on XML: the first stores information required to connect multiple data sources and retrieve clinical data, and the second provides a data architecture to acquire clinical and repository data. All data were anonymized and encrypted using the Advanced Encryption Standard. X.509 certificates were applied to secure data access. Furthermore, we tested the feasibility of implementing these models in the information management system for biobanking. The data concept can provide clinical and repository data of biospecimens. Only authorized receivers can access these data. Sensitive and personal data are not accessible by the data receivers. The data receiver cannot backtrack to the individual donor using the data model. The acquired data can be converted into a text file format supported by familiar statistical software. Supplementary tools were implemented to generate and view XML documents based on these data models. This data model provides an effective approach to distribute clinical and repository data from different data sources to enable data analysis compliant with data privacy regulations.

Tumor detection under cystoscopy with transformer-augmented deep learning algorithm.

Objective.Accurate tumor detection is critical in cystoscopy to improve bladder cancer resection and decrease recurrence. Advanced deep learning algorithms hold the potential to improve the performance of standard white-light cystoscopy (WLC) in a noninvasive and cost-effective fashion. The purpose of this work is to develop a cost-effective, transformer-augmented deep learning algorithm for accurate detection of bladder tumors in WLC and to assess its performance on archived patient data.Approach.'CystoNet-T', a deep learning-based bladder tumor detector, was developed with a transformer-augmented pyramidal CNN architecture to improve automated tumor detection of WLC. CystoNet-T incorporated the self-attention mechanism by attaching transformer encoder modules to the pyramidal layers of the feature pyramid network (FPN), and obtained multi-scale activation maps with global features aggregation. Features resulting from context augmentation served as the input to a region-based detector to produce tumor detection predictions. The training set was constructed by 510 WLC frames that were obtained from cystoscopy video sequences acquired from 54 patients. The test set was constructed based on 101 images obtained from WLC sequences of 13 patients.Main results.CystoNet-T was evaluated on the test set with 96.4 F1 and 91.4 AP (Average Precision). This result improved the benchmark of Faster R-CNN and YOLO by 7.3 points in F1 and 3.8 points in AP. The improvement is attributed to the strong ability of global attention of CystoNet-T and better feature learning of the pyramids architecture throughout the training. The model was found to be particularly effective in highlighting the foreground information for precise localization of the true positives while favorably avoiding false alarmsSignificance.We have developed a deep learning algorithm that accurately detects bladder tumors in WLC. Transformer-augmented AI framework promises to aid in clinical decision-making for improved bladder cancer diagnosis and therapeutic guidance.

Conceptual Framework and Documentation Standards of Cystoscopic Media Content for Artificial Intelligence.

BACKGROUND: The clinical documentation of cystoscopy includes visual and textual materials. However, the secondary use of visual cystoscopic data for educational and research purposes remains limited due to inefficient data management in routine clinical practice. METHODS: A conceptual framework was designed to document cystoscopy in a standardized manner with three major sections: data management, annotation management, and utilization management. A Swiss-cheese model was proposed for quality control and root cause analyses. We defined the infrastructure required to implement the framework with respect to FAIR (findable, accessible, interoperable, re-usable) principles. We applied two scenarios exemplifying data sharing for research and educational projects to ensure the compliance with FAIR principles. RESULTS: The framework was successfully implemented while following FAIR principles. The cystoscopy atlas produced from the framework could be presented in an educational web portal; a total of 68 full-length qualitative videos and corresponding annotation data were sharable for artificial intelligence projects covering frame classification and segmentation problems at case, lesion and frame levels. CONCLUSION: Our study shows that the proposed framework facilitates the storage of the visual documentation in a standardized manner and enables FAIR data for education and artificial intelligence research.

PlexusNet: A neural network architectural concept for medical image classification.

State-of-the-art (SOTA) convolutional neural network models have been widely adapted in medical imaging and applied to address different clinical problems. However, the complexity and scale of such models may not be justified in medical imaging and subject to the available resource budget. Further increasing the number of representative feature maps for the classification task decreases the model explainability. The current data normalization practice is fixed prior to model development and discounting the specification of the data domain. Acknowledging these issues, the current work proposed a new scalable model family called PlexusNet; the block architecture and model scaling by the network's depth, width, and branch regulate PlexusNet's architecture. The efficient computation costs outlined the dimensions of PlexusNet scaling and design. PlexusNet includes a new learnable data normalization algorithm for better data generalization. We applied a simple yet effective neural architecture search to design PlexusNet tailored to five clinical classification problems that achieve a performance noninferior to the SOTA models ResNet-18 and EfficientNet B0/1. It also does so with lower parameter capacity and representative feature maps in ten-fold ranges than the smallest SOTA models with comparable performance. The visualization of representative features revealed distinguishable clusters associated with categories based on latent features generated by PlexusNet. The package and source code are at https://github.com/oeminaga/PlexusNet.git.

Artificial Intelligence Reveals Distinct Prognostic Subgroups of Muscle-Invasive Bladder Cancer on Histology Images.

Muscle-invasive bladder cancer (MIBC) is a highly heterogeneous and costly disease with significant morbidity and mortality. Understanding tumor histopathology leads to tailored therapies and improved outcomes. In this study, we employed a weakly supervised learning and neural architecture search to develop a data-driven scoring system. This system aimed to capture prognostic histopathological patterns observed in H&E-stained whole-slide images. We constructed and externally validated our scoring system using multi-institutional datasets with 653 whole-slide images. Additionally, we explored the association between our scoring system, seven histopathological features, and 126 molecular signatures. Through our analysis, we identified two distinct risk groups with varying prognoses, reflecting inherent differences in histopathological and molecular subtypes. The adjusted hazard ratio for overall mortality was 1.46 (95% CI 1.05-2.02; z: 2.23; p = 0.03), thus identifying two prognostic subgroups in high-grade MIBC. Furthermore, we observed an association between our novel digital biomarker and the squamous phenotype, subtypes of miRNA, mRNA, long non-coding RNA, DNA hypomethylation, and several gene mutations, including FGFR3 in MIBC. Our findings underscore the risk of confounding bias when reducing the complex biological and clinical behavior of tumors to a single mutation. Histopathological changes can only be fully captured through comprehensive multi-omics profiles. The introduction of our scoring system has the potential to enhance daily clinical decision making for MIBC. It facilitates shared decision making by offering comprehensive and precise risk stratification, treatment planning, and cost-effective preselection for expensive molecular characterization.

Efficient Augmented Intelligence Framework for Bladder Lesion Detection.

PURPOSE: Development of intelligence systems for bladder lesion detection is cost intensive. An efficient strategy to develop such intelligence solutions is needed. MATERIALS AND METHODS: We used four deep learning models (ConvNeXt, PlexusNet, MobileNet, and SwinTransformer) covering a variety of model complexity and efficacy. We trained these models on a previously published educational cystoscopy atlas (n = 312 images) to estimate the ratio between normal and cancer scores and externally validated on cystoscopy videos from 68 cases, with region of interest (ROI) pathologically confirmed to be benign and cancerous bladder lesions (ie, ROI). The performance measurement included specificity and sensitivity at frame level, frame sequence (block) level, and ROI level for each case. RESULTS: Specificity was comparable between four models at frame (range, 30.0%-44.8%) and block levels (56%-67%). Although sensitivity at the frame level (range, 81.4%-88.1%) differed between the models, sensitivity at the block level (100%) and ROI level (100%) was comparable between these models. MobileNet and PlexusNet were computationally more efficient for real-time ROI detection than ConvNeXt and SwinTransformer. CONCLUSION: Educational cystoscopy atlas and efficient models facilitate the development of real-time intelligence system for bladder lesion detection.

Real-time Detection of Bladder Cancer Using Augmented Cystoscopy with Deep Learning: a Pilot Study.

BACKGROUND: Detection of bladder tumors under white light cystoscopy (WLC) is challenging yet impactful on treatment outcomes. Artificial intelligence (AI) holds the potential to improve tumor detection; however, its application in the real-time setting remains unexplored. AI has been applied to previously recorded images for post hoc analysis. In this study, we evaluate the feasibility of real-time AI integration during clinic cystoscopy and transurethral resection of bladder tumor (TURBT) on live, streaming video. METHODS: Patients undergoing clinic flexible cystoscopy and TURBT were prospectively enrolled. A real-time alert device system (real-time CystoNet) was developed and integrated with standard cystoscopy towers. Streaming videos were processed in real time to display alert boxes in sync with live cystoscopy. The per-frame diagnostic accuracy was measured. RESULTS AND LIMITATIONS: Real-time CystoNet was successfully integrated in the operating room during TURBT and clinic cystoscopy in 50 consecutive patients. There were 55 procedures that met the inclusion criteria for analysis including 21 clinic cystoscopies and 34 TURBTs. For clinic cystoscopy, real-time CystoNet achieved per-frame tumor specificity of 98.8% with a median error rate of 3.6% (range: 0 - 47%) frames per cystoscopy. For TURBT, the per-frame tumor sensitivity was 52.9% and the per-frame tumor specificity was 95.4% with an error rate of 16.7% for cases with pathologically confirmed bladder cancers. CONCLUSIONS: The current pilot study demonstrates the feasibility of using a real-time AI system (real-time CystoNet) during cystoscopy and TURBT to generate active feedback to the surgeon. Further optimization of CystoNet for real-time cystoscopy dynamics may allow for clinically useful AI-augmented cystoscopy.