(UroToday.com) The 2024 IBCN annual meeting included a session on emerging technologies in bladder cancer, featuring a presentation by Dr. Joseph Liao discussing augmented cystoscopy with deep learning. Dr. Liao notes that early diagnosis and accurate staging for bladder cancer is critical, with current strategies for risk stratification based on clinicopathological factors. Precision oncology relies on diagnostics, surgery, and therapy. Artificial intelligence is an enabling tool for clinical decision support and integrated risk stratification.
Dr. Liao then discussed a case of a 75 year old male, former smoker (12 pack-year), who presented with gross hematuria. A CT-IVP showed “incomplete bladder distention, 1.9 x 1.4 x 1.7 cm rounded focus at the left lateral bladder wall,” with office cystoscopy confirming a bladder tumor. This patient subsequently underwent a TURBT, staged as high grade Ta, followed by BCG, and an in office cystoscopy demonstrated recurrence of high grade T1, followed by another TURBT in the OR. Most likely, this story continues for months/years with continued intravesical therapy and TURBTs. Dr. Liao questions whether there is a way for artificial intelligence to support this clinical work flow?
Dr. Liao emphasized that there are recognized short comings for cystoscopy and TURBT, including operator dependency, missed tumors/multifocal disease, and indeterminate lesions/CIS. TURBT quality is critical and deceptively difficult, leading to incomplete resection, under-staging, and requiring repeat TURBT. There are enhanced cystoscopy techniques, including blue light cystoscopy and other emerging technologies, but there are considerations for technology access, cost effectiveness, and clinical workflow. Finally, clinical documentation is suboptimal based on text from cystoscopy notes, operative reports, intravesical therapy treatment notes, and pathology reports. Several examples of enhanced cystoscopy technologies are as follows:
As such, how may artificial intelligence improve TURBT and enhanced imaging? In 2019, Dr. Liao’s group developed a deep learning algorithm for augmented cystoscopic detection of bladder cancer. Using video frames containing histologically confirmed papillary urothelial carcinoma that were manually annotated, Shkolyar et al.1 constructed CystoNet, an image analysis platform based on convolutional neural networks, for automated bladder tumor detection. In the validation dataset, per-frame sensitivity and specificity were 90.9% (95% CI, 90.3-91.6%) and 98.6% (95% CI, 98.5-98.8%), respectively. Per-tumor sensitivity was 90.9% (95% CI, 90.3-91.6%). Representative bladder cancer detection using CystoNet is as follows:
The following algorithm provided by Dr. Liao2 highlights an overview of artificial intelligence-enhanced cystoscopy for precise tumor detection:
From a work flow standpoint, Dr. Liao’s group has run a pilot study evaluating the feasibility of real-time artificial intelligence integration during clinic cystoscopy and TURBT on live, streaming video.3 Real-time CystoNet was successfully integrated into the operating room during TURBT and clinic cystoscopy in 50 consecutive patients. 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.
While the classical papillary tumors may be “easy” to annotate, Dr. Liao’s group has continued development of their technology to harness annotation of flat lesions using a computer vision annotation tool:
This can also be augmented by blue light cystoscopy:
Several goals highlighted by Dr. Liao are to provide a cystoscopy imaging dataset, create 3D bladder reconstruction using standard white light cystoscopy, and provide additional assessment of the tumor resection bed. Dr. Liao’s group has also partnered with Convergent Genomics to integrate cystoscopy imaging data with urine minimal residual disease detection assessing variant types in the TURBT specimens:
Dr. Liao concluded his presentation by discussing augmented cystoscopy with deep learning with the following take-home points:
- Augmented cystoscopy holds promising potential for clinical detection support for tumor detection
- Well annotated dataset critical, including CIS and cancer-mimicking benign lesions
- Ongoing work to optimize AI algorithms using video sequences for training, real time clinic/OR integration, and user interface
- Resection bed characterization and blue light cystoscopy can be enhanced with artificial intelligence
- Multimodal models are needed that integrate clinical history, liquid biopsy, biomarkers, pathology, and HER
- There are opportunities for multicenter collaboration
Presented by: Joseph Liao, MD, Stanford University School of Medicine, Palo Alto, CA
Written by: Zachary Klaassen, MD, MSc – Urologic Oncologist, Associate Professor of Urology, Georgia Cancer Center, Wellstar MCG Health, @zklaassen_md on Twitter during the 2024 International Bladder Cancer Network (IBCN) Annual Meeting, Bern, Switzerland, Thurs, Sept 19 – Sat, Sept 21, 2024
References:
- Shkolyar E, Jia X, Chang TC, et al. Augmented bladder tumor detection using deep learning. Eur Urol. 2019 Dec;76(6):714-718.
- Shkolyar E, Zhou SR, Carlson CJ, et al. Optimizing cystoscopy and TURBT: Enhanced imaging and artificial intelligence. Nat Rev Urol. 2024 Jul 9 [Epub ahead of print].
- Chang TC, Shkolyar E, Del Giudice F, et al. Real-time detection of bladder cancer using augmented cystoscopy with deep learning: A pilot study. J Endourol. 2023 Jul 11 [Epub ahead of print].