Disclaimer: Unfold AI was previously known as iQuest
A high-level summary of documented efforts to validate Avenda Health’s iQuest software, October 2022
Avenda Health’s iQuest software is an artificial intelligence (AI)-based decision support tool, for use in conjunction with medical imaging and biopsy data to plan prostate cancer interventions. iQuest supports the prostate oncological workflow by helping users segment MR image features, visualize cancer risk in 3D, and select sites for treatment. To date various retrospective studies of clinical data have been used to validate iQuest features, including multiple peer-reviewed abstracts and several manuscripts in preparation. These validation efforts are summarized below.
AI Prostate Segmentation Algorithm (AIPSA)
Avenda Health developed an AI algorithm for automatic segmentation of the prostate based on MRI images and validated that algorithm on 137 multi-institutional test cases. The algorithm had excellent performance across a wide range of prostate volumes, scanner models, and image resolutions, with a mean dice similarity coefficient of 0.93. Furthermore, during head-to-heat comparisons against the prostate segmentation feature of a competing product (In Vivo DynaCAD), the Avenda algorithm performed best with significant improvements to dice similarity (0.93 vs 0.86) and relative volume error (4.9% vs 15.2%).
AI Lesion Contouring Algorithm
Avenda Health developed an AI algorithm to aid users in defining a lesion contour for use during prostate interventions such as targeted therapy. The core feature of this algorithm is the cancer estimation map (CEM), a 3D map of estimated clinically significant cancer (csPCa) risk throughout the prostate. The CEM was developed using multi-institutional fusion biopsy data and validated on 100 multi-institutional prostatectomy cases [1]. The CEM was proven to be highly effective at predicting csPCa, with a mean AUC of 0.92 and balanced accuracy of 85%. Furthermore, using 50 prostatectomy cases from an external population, default lesion contours derived from the CEM outperformed conventional planning [1]. Avenda lesion contours had greater csPCa sensitivity and a higher rate of index tumor encapsulation (97% and 90% respectively) than the original radiologist-defined target (37%, 0%), hemi-gland margins (94%, 66%) and 1-cm uniform margins (93%, 82%). In the same study, iQuest’s encapsulation confidence score feature was able to accurately predict the rate of csPCa encapsulation, with a R2 value of 0.98 and a median error of only 4%. Avenda LCA validation efforts have been published in 5 peer-reviewed and award winning abstracts [2-6], and a manuscript is being prepared for journal submission [1].
Physician Study of Lesion Contouring
In addition to internal studies confirming the accuracy and efficacy of user interactions with iQuest as a whole, the lesion contouring task was evaluated via a multi-reader (N = 10) multi-case (N = 50) study using radical proctectomy specimens as ground truth [7-10]. Readers (physicians) using iQuest produced lesion contours that identified more csPCa than manually defined contours (97% vs 38%), while sparing more benign tissue than hemi-gland contours (72% vs 53%). The balanced accuracy of iQuest contours was significantly higher than both comparator methods, and reading time was reduced by 40%. A manuscript is being prepared for journal submission [9].
Avenda Health Automatic Ablation Site Placement (AASP)
Avenda Health developed an algorithm to aid users in defining ablation sites that completely encapsulate an arbitrary lesion contour [9]. The algorithm functioned across a wide range of ablation shapes and sizes, dramatically reduced the time required to place ablation sites (from >10 minutes to 2 seconds), and improved lesion contour coverage compared to manual methods (94% vs 87%).
References
[1] Priester AM, Fan R, Shubert J, et al.: “Prediction and Visualization of Lesion Extent in Intermediate Risk Prostate Cancer using Artificial Intelligence” Manuscript Draft intended for publication in the Journal of Urology, October 2022
[2] Priester A, Zhou S, Kamothi Y, et al: “Improving Prostate Cancer Margin Prediction with Machine Learning. In: Engineering in Urology Society.” Engineering in Urology Society 33rd Annual Meeting, Chicago 2019; pp 45.
[3] Alan P, Richard F, Joshua S, et al: “Defining Prostate Cancer Focal Therapy Treatment Margins with a Machine Learning Model: Improvement Upon Hemi-Gland Ablation.” Engineering in Urology Society 35th Annual Meeting, New Orleans 2022; pp 100.
[4] Alan P, Richard F, Joshua S, et al: “A Novel AI-Derived Metric Identifies Favorable Candidates for Focal Therapy of Prostate Cancer and Accurately Predicts Treatment Margin Efficacy.” Engineering in Urology Society 35th Annual Meeting, New Orleans 2022; pp 50.
[5] Alan P, Richard F, Joshua S, et al: “A Validated Prostate Cancer Probability Map To Aid In Focal Treatment Planning.” AdMeTech Foundation Sixth Global Summit on Precision Diagnosis and Treatment of Prostate Cancer, 2022.
[6] Alan P, Richard F, Joshua S, et al: “A Novel AI-Derived Metric Identifies Optimal Candidates For Focal Therapy Of Prostate Cancer And Accurately Predicts Treatment Margin Efficacy” AdMeTech Foundation Sixth Global Summit on Precision Diagnosis and Treatment of Prostate Cancer, 2022.
[7] “22-PR-1001 AH Lesion Contour Algorithm Reader Study Protocol Revision,” Avenda Health Internal Document.
[8] “22-RP-1001 AH Lesion Contour Algorithm Reader Study Report,” Avenda Health Internal Document.
[9] Priester AM, Mota S, Shubert S, “A multi-reader, multi-case study of AI-informed prostate cancer lesion contouring” Manuscript Draft intended for publication in European Urology, October 2022
[10] Priester AM, Fan R, Bhatnagar P, et al: ” Automatic Ablation Site Selection For Focal Therapy.” Engineering in Urology Society 34th Annual Meeting, Washington DC 2021; pp 82
Other Supporting Publications for iQuest
- Magnetic Resonance Imaging Underestimation of Prostate Cancer Geometry: Use of Patient Specific Molds to Correlate Images with Whole Mount Pathology
- Prostate Cancer Detection With Magnetic Resonance-Ultrasound Fusion Biopsy: The Role of Systematic and Targeted Biopsies
- Utility for mutiparametric MRI for Predicting Residual Clinically Significant Prostate Cancer After Focal Laser Ablation
- Focal Boost to the Intraprostatic Tumor in External Beam Radiotherapy for Patients With Localized Prostate Cancer: Results From the FLAME Randomized Phase III Trial
