SOFTWARE OPTIMIZATION & VALIDATION STUDIES
ECDP Poster Presentation
Developing an optimal artificial intelligence-based logistic model for automated bladder cancer screening in digital urine cytology
Resources and News
Educational resources and news from AIxMed to keep you informed
and up-to-date on our AI empowered digital cytology solutions.
and up-to-date on our AI empowered digital cytology solutions.
.png)
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
AIxTHY THYROID FINE NEEDLE ASPIRATION CLINICAL STUDIES
October 2025
Pathology Visions Oral Presentation
COMING SOON - Redefining Thyroid Cytology with AIxTHY: Improved Accuracy and Efficiency in FNAC Interpretation
Pathology Visions Oral Presentation
COMING SOON - Redefining Thyroid Cytology with AIxTHY: Improved Accuracy and Efficiency in FNAC Interpretation
STUDY DESIGN:
RESULTS:
CONCLUSIONS:
AIxURO URINE CYTOLOGY CLINICAL STUDIES
May 2025
ICC 2025 Oral Presentation
AIxURO: Enhancing Diagnostic Accuracy in Bladder Caner Detection with AI-Assisted Urine Cytology
ICC 2025 Oral Presentation
AIxURO: Enhancing Diagnostic Accuracy in Bladder Caner Detection with AI-Assisted Urine Cytology
STUDY DESIGN:
- Retrospective study using 200 ThinPrep urine cytology slides.
- Ground truth: NHGUC (n=100), AUC (n=35), SHGUC (n=32), HGUC (n=33).
- Slides digitized using: (1) Mikroscan SLxCyto( Mikroscan-WSIs); (2) Customized Huron scanner (Huron-WSIs).
- WSIs were analyzed using AIxURO, a disease-specific AI algorithm for bladder cancer.
- Three-arm blinded review by 1 cytopathologist and 2 cytologists, with 2-week washout periods:
- Arm 1: Conventional microscopy
- Arm 2: AIxURO with Mikroscan-WSIs
- Arm 3: AIxURO with Huron-WSIs
- Performance Metrics:
- Comparison of study diagnosis with urine cytology ground truth diagnosis (n=200) for 3 arms, using following threshold: AUC+(AUC, SHGUC and HGUC) cases as positive; NHGUC cases as negative.
- Assessment of agreement between study diagnoses and surgical pathology-confirmed bladder cancer (n=72) among the three arms.
- Evaluation of bladder cancer prediction in hematuria-indicated cases (n=16) across the three arms.
RESULTS:
- Urine cytology ground truth cases: Arm 2 and Arm3 exhibited higher sensitivity (85.0% and 88.3%) compared to Arm 1 (79.3%) with comparable accuracy across all arms (Arm 1: 86.8%, Arm 2: 85.3%, Arm 3: 85.3%).
- Biopsy-confirmed bladder cancer cases: Arm 2 and Arm 3 showed improved sensitivity (92.0% and 93.2%) and accuracy (82.9% and 82.4%) compared to Arm 1 (84.6% sensitivity, 78.2% accuracy).
- Hematuria cases: Arms 2 and 3 achieved superior sensitivity (96.7% and 100.0%) and accuracy (89.6% and 91.7%) compared to Arm 1 (90.0% sensitivity, 85.4% accuracy).
CONCLUSIONS:
AIxURO consistently enhanced diagnostic sensitivity and maintained accuracy across multiple scanners, outperforming microscopy in detecting bladder cancer. It demonstrated strong predictive performance in high-risk subgroups, supporting its utility in real-world clinical workflows and digital cytology integration.
AI SOFTWARE DEVELOPMENT
May 2025
ICC 2025 - Oral Presentation
Does AI performance in Thyroid Cancer Diagnosis Differ Between Single-Layer and Multi-Z-Stacked Imaging?
ICC 2025 - Oral Presentation
Does AI performance in Thyroid Cancer Diagnosis Differ Between Single-Layer and Multi-Z-Stacked Imaging?
STUDY DESIGN:
- 71 ThinPrep thyroid FNAC slides selected by consensus cytology diagnosis
- Cases included: TBS-II (n=35), TBS-IV (n=6), TBS-VI (n=30), confirmed by biopsy
- Slides digitized using 3DHistech scanner to create paired whole-slide images:
- S-WSI: single-layer WSI
- 7-WSI: seven-layer Z-stacked WSI
- AI-assisted review used an AIxTHY model to detect cancer cells and guide interpretation
- Three cytologists independently reviewed both S-WSI and 7-WSI sets
- Total of 213 reads (71 cases × 3 reviewers) analyzed
- Two diagnostic threshold:
- Threshold 1: Positive = TBS-V/VI; Negative = TBS-II
- Threshold 2: Positive = TBS-IV/V/VI; Negative = TBS-II
- Outcomes measured:
- Binary diagnostic sensitivity and specificity (under both thresholds)
- Agreement with consensus diagnoses
- Interobserver agreement (Cohen’s κ)
RESULTS:
- Threshold 1 (TBS-V/VI vs. TBS-II):
- Sensitivity: 7-WSI 88.9% vs. S-WSI 81.1%
- Specificity: 7-WSI 94.3% vs. S-WSI 96.2% (not significant)
- Threshold 2 (TBS-IV/V/VI vs. TBS-II):
- Sensitivity: 7-WSI 86.1% vs. S-WSI 80.6% (p < 0.05)
- Specificity: 7-WSI (%): All cases: 7-WSI 88.6% vs. S-WSI 91.4%
- Consensus Agreement (%): All cases: 7-WSI 74.6%vs. S-WSI 61.5%
- TBS-VI: 65.6% vs. 36.7%
- TBS-IV: 44.4% vs. 27.8%
- TBS-II: 87.6% vs. 88.6%
- Interobserver Agreement (Cohen’s κ): Overall: 0.366 vs. 0.351
- TBS-IV: 7-WSI 0.189 vs. S-WSI 0.075
- TBS-VI: 0.420 vs. 0.319
- TBS-II: 0.549 vs. 0.604
CONCLUSIONS:
Seven-layer Z-stacked WSIs (7-WSI) enhanced AI-assisted thyroid cancer diagnosis and improved interobserver agreement among cytologists, especially in indeterminate (TBS-IV) and malignant (TBS-VI) cases, supporting their clinical value over single-layer WSIs.
June 2024
Developing an optimal artificial intelligence-based logistic model for automated bladder cancer screening in digital urine cytology
ECDP 2024 Poster Presentation Receiver-Operator Curve (ROC) study to assess AIxURO logistic model’s accuracy to predict bladder cancer
STUDY DESIGN:
- 1856 urine cytology cases (1466 negative and 390 positive)- AI training set
- 169 urine cytology cases (88 negative, 81 positive)- Validation set
- AIxURO classifies abnormal urothelial cells into 2 categories based on The Paris System 2.0: “Suspicious” (SHGUC or HGUC) and “Atypical “(AUC), with the final interpretation deferred to a pathologist
- Logistic regression performed to predict presence of cancer, including variables such as total # suspicious cells, total # atypical cells, and predictive accuracy using sensitivity and specificity
RESULTS:
- Optimal performance of the training set (based on the total number of atypical cells) was 10 cells (cytospin) and 49 cells (CytoRich)
- Training Set Sensitivity and Specificity: 75.9% and 73.0%
- Validation Set Sensitivity and Specificity: 75.3% and 87.5%
CONCLUSIONS:
The logistic model supports the optimal cut-off values of at least 10 cells (cytospin) and 49 cells (CytoRich) for the number of atypical cells required for a high concordance with bladder cancer as the final outcome.
April 2024
SOFTWARE OPTIMIZATION & VALIDATION STUDIES
Academic Article: Acta Cytologica
Evaluating urine cytology slide digitization efficiency: A comparative study using an artificial intelligence-based heuristic scanning simulation and multiple Z-plane scanning
Evaluating urine cytology slide digitization efficiency: A comparative study using an artificial intelligence-based heuristic scanning simulation and multiple Z-plane scanning
Hang JF, Ou YC, Yang WL et al. Acta Cytol. 2024;68(4):342-350. doi: 10.1159/000538985. Epub 2024 Apr 22. PMID: 38648759. AIxMED Technical Review Study, Stage II (Nat’l Yang Ming Chiao Tung Univ, VGHTPE, Tung Univ, Tung’s Taichung MetroHarbor)
STUDY DESIGN:
- 52 urine cytology slides (cytospin, ThinPrep, and CytoRIch) scanned with 21 Z-plane and a heuristic scan simulation method to generate whole slide images (WSI) using a Leica Aperio AT2 scanner
- An AI algorithm inferred the WSI from 21 Z-planes to quantitate total number of cells suspicious for high grade urothelial carcinoma (SHGUC) / high-grade urothelial carcinoma (HGUC)=[SHGUC+]
- The heuristic scan simulation calculated the total number of SHGUC+ using the 21 Z-plane scan data
- Performance metrics evaluated were SHGUC+ cell coverage rates, scanning times, file size, and AI-aided interpretation of WSI compared to the original cytology diagnosis for the 21 Z-plane scan and the heuristic scan
RESULTS:
- SHGUC+ Coverage Rates: Heuristic scanning coverage rates were similar to 5 Z-plane scans for all 3 preparation types (0.78 to 0.91 vs 0.75 to 0.88; p = 0.451 to 0.578)
- Scanning Time: Heuristic scanning significantly reduced scanning time (137.2 to 635.0 seconds vs 332.6 to 1,278.8 seconds; p < 0.05)
- Image File Size: Heuristic scanning significantly reduced image file size (0.51 to 2.10 GB vs. 1.16 to 3.10 GB; p < 0.05)
- AI-aided Interpretation: Heuristic scanning had higher rates of accurate interpretation compared to single Z-plan scanning (62.5% vs. 37.5%)
CONCLUSIONS:
Heuristic scanning showed improved scanning times and AI-aided cytologic interpretation while reducing overall image file size and maintaining similar scanning coverage for SHGUC+ cells for 3 urine cytology preparation types.
March 2024
AIxURO URINE CYTOLOGY CLINICAL STUDIES
USCAP Poster Presentation
Evaluating the feasibility of a two-step workflow in artificial intelligence-assisted digital urine cytology
Evaluating the feasibility of a two-step workflow in artificial intelligence-assisted digital urine cytology
USCAP 2024 Poster Presentation / Abstract Wei-Lei Yang, PhD Show Chwan Hospital, clinical pilot study
STUDY DESIGN:
- 116 urine cytology slides (76 Cytospin, 40 CytoRich)
- Consensus diagnosis by two senior cytopathologists for ground truth, using TPS2.0, resulting in
- 30 positive slides (AUC/SHGUC/HGUC)
- 86 negative slides (NHGUC)
- Consensus diagnosis by two senior cytopathologists for ground truth, using TPS2.0, resulting in
- 3-Arm study with 1 cytopathologist (CP) and 2 cytologist (CT) reviewers and a 2-week washout period between each arm; and analysis of the cytopathologist paired with one of the cytologists to mimic clinical practice of cytologist review and referral to cytopathologist
- Arm 1: Microscopy
- Arm 2: Digital whole slide image review
- Arm 3: Digital image review using artificial intelligence software (AIxURO)
- Performances Metrics: Sensitivity, specificity, PPV, NPV, accuracy and total diagnostic (review) time
RESULTS:
- Sensitivity: Improved with AI-assistance for CP+CTA (90% vs 76.7%) and CP+CTB (76.7% vs 76.7%) compared with microscopy (but not with WSI review alone)
- Negative Predictive Value: Improved with AI-assistance for CP+CTA (96.4% vs 92.2%) and CP+CTB (92% vs 92.3%) compared with microscopy (but not with WSI review alone)
- Specificity: Decreased in AI-assistance for CP+CTA (93% vs 96.5%) and CP+CTB (92% vs 92.3%) compared to microscopy alone
- Positive Predictive Value: Decreased in AI-assistance for CP+CTA (81.8% vs 88.5%) and CP+CTB (79.3% vs 92%) compared to microscopy alone
- Overall, Arm 2 (WSI review) showed no improvement compared to microscopy for either CP+CTA or CP+CTB
- Overall Review Time: AI-assisted review decreased the total review time (72.2 min and 110.4 min), compared to microscopy (210.2 min and 244.7 min), whereas WSI review took as long or longer (227.1 min and 243.8 min) than microscopy
CONCLUSIONS:
AI-assisted urine cytology review markedly reduces review time while increasing sensitivity and NPV using TPS2.0 compared to conventional microscopy. However, specificity, PPV and accuracy are slightly diminished. Pairing a cytopathologist (CP) with different cytologists (CT) also influences the diagnostic outcomes and metrics.
March 2024
AIxURO URINE CYTOLOGY CLINICAL STUDIES
USCAP Oral Platform l Presentation
Assessing the clinical practicability of artificial intelligence assisted reporting using the Paris System for urinary cytology
Assessing the clinical practicability of artificial intelligence assisted reporting using the Paris System for urinary cytology
USCAP 2024 Oral Platform l Presentation- Tien-Jen Liu, MD Show Chwan Hospital clinical pilot study
STUDY DESIGN:
- 116 urine cytology slides with consensus diagnosis (ground truth) by a panel of experts to 86 NHGUC, 12 AUC, 11 SHGUC, and 7 HGUC, scanned with Leica Aperio AT2 to create a whole slide image (WSI)
- 1 Cytopathologist and 2 cytologists reviewed all slides/images in each arm independently, recording the time required to diagnosis and The Paris System (TPS) cytologic diagnosis, with a 2-week washout period between each review
- Arm 1: Microscopic review of the cytology slide
- Arm 2: Review of the scanned whole slide image (WSI)
- Arm 3: Review of images with AI-assistance software (AIxURO)
- Metrics: TPS diagnostic category compared with the expert panel and Total time spent on review
RESULTS:
- NHGUC, SHGUC and HGUC: AIxURO showed higher specificity than microscopy, but lower sensitivity
- AUC: AIxURO showed higher sensitivity but lower specificity than microscopy
- The performance of Arm 2 (using the whole slide digital image only for analysis) was poorest overall, compared to the other 2 arms
- There were performance inconsistencies between reviewers. For example, sensitivity for SHGUC was decreased for the pathologist and one cytologist, whereas the other cytologist noted an increase in sensitivity. One cytologist had a decrease in sensitivity for HGUC compared to the other 2 reviewers.
- AIxURO showed the largest reduction in time spent on review (32-45% less for the pathologist and 10-62% for the cytologists). The cytopathologist took the longest time to review AUC and the shortest for HGUC, whereas the cytologists took the most time to review SHGUC and the least on NHGUC.
CONCLUSIONS:
AI-assisted AIxURO outperformed microscopy in diagnostic accuracy for AUC while maintaining comparable accuracy across other TPS categories and significantly reducing total review time for all reviewers. WSI review alone did not improve diagnostic accuracy or efficiency.
November 2023
WHOLE SLIDE IMAGE SCANNER STUDIES
Academic Article: Journal of Pathology Informatics
Comparative evaluation of slide scanners, scan settings, and cytopreparations for digital urine cytology
Comparative evaluation of slide scanners, scan settings, and cytopreparations for digital urine cytology
Hang JF, Ou YC, Yang WL et al. J Pathol Inform. 2023 Nov 4;15:100346. Doi: 10.1016/j.jpi.2023.100346. PMID: 38125926; PMCID: PMC10730371. AIxMED Technical Review Study, Stage I
STUDY DESIGN:
- 14 urine cytology specimens, equally divided and prepped into CytoSpin and ThinPrep specimens, were scanned for WSI by Leica Aperio AT2 and Hamamatsu NanoZoom S360 at 3 focus modes (default, semiautomatic, and manual) for single layer scanning and with a manual focus mode for 21 Z-layer scanning
- Performance metrics evaluated included scanning success rate, AI-algorithm-inferred atypical cell numbers and coverage rates (atypical cells in single or multiple Z-layers divided by total atypical cells), scanning time, and image file size
RESULTS:
Success Scanning Rates
- Default Mode Scanning: 85.7% (Cytospin, Leica) and 92.9% (ThinPrep, Leica; Cytospin and ThinPrep, Hamamatsu) success rates
- Semi-Auto Mode Scanning: 92.9% (Cytospin, Leica; Cytospin, Hamamatsu and ThinPrep, Hamamatsu) or 100% (ThinPrep, Leica) success
- Manual Scanning: 100% success (all preps with all scanners)
Scanning Times (median)
- Cytospin, Leica: 1010 seconds
- Cytospin, Hamamatsu: 275 seconds
- ThinPrep, Leica: 5357 seconds
- ThinPrep, Hamamatsu: 1429 seconds
File Image Sizes of WSI (using manual focus mode and 21 Z-layer scan settings)
- Cytospin, Leica: 2.0 GB
- Cytospin, Hamamatsu: 4.4 GB
- ThinPrep, Leica: 13.1 GB
- ThinPrep, Hamamatsu: 25.5 GB
CONCLUSIONS:
Semi-automatic and manual focus modes provide the most successful slide scanning with rates up to 100%. A minimum of 9-layer Z-stacking at 1 µm intervals is necessary to cover 80% of atypical cells. Cytospin preparations take less time to scan and result in smaller file sizes than do ThinPrep slides, regardless of the scanner. Z-stacking enhances AI inferred quality and coverage rates of atypical cells but with longer scanning times and larger image files.
November 2023
WHOLE SLIDE IMAGE SCANNER STUDIES
ASC Poster Presentation
Evaluating image quality and reproducibility of digital scanners for whole-slide imaging in urine cytology slides for artificial intelligence applications
Evaluating image quality and reproducibility of digital scanners for whole-slide imaging in urine cytology slides for artificial intelligence applications
ASC 2023 Poster Presentation / Abstract (JASC 2023; Vol12(5):S74-75). Yang WL, Hang JF, Ou YC et al.
STUDY DESIGN:
- 20 urine cytology slides, 5 types of preparation (cytospin, ThinPrep nonGYN, ThinPrep Urocyte, and BD CytoRich) were digitized using 3 digital scanners (Roch DP200, Roch DP600, and Hamamatsu Nanozoomer 360)
- Images were evaluated for quality, focus and color from the default, manual and advanced scanning modes by a senior cytologist
RESULTS:
- Roche DP200 and DP600 scanners achieved good quality WSI (30-50% in default mode and 40-65% in manual mode)
- Hamamatsu achieved 90% good quality imaging only in the manual mode, whereas the default quality dropped to 15%. It also had a lower coefficient of variation for average number of atypical cells across all preparation types
CONCLUSIONS:
Overall, Hamamatsu scanner showed better performance for image quality in the manual mode than Roche scanners, which performed better in the default mode. Hamamatsu scanners also had improved detection of atypical urothelial cells among triplicate images from the same samples compared to Roche scanners.
November 2023
AIxURO URINE CYTOLOGY CLINICAL STUDIES
ASC Oral Presentation
Evaluating the feasibility of a deep learning-based atypical cell gallery for clinical diagnosis in urine cytology
Evaluating the feasibility of a deep learning-based atypical cell gallery for clinical diagnosis in urine cytology
ASC 2023 Oral Platform/ Abstract (JASC 2023; Vol 12(5):S51-52) Liu TJ, Yang WC, Huang SM, et al.
STUDY DESIGN:
- 116 urine cytology slide scanned as WSI and analyzed by AI deep-learning software system and ranked as most suspicious or atypical urothelial cells based on The Paris System 2.0 (TPS 2.0) for Reporting Urine Cytology
- Cell image gallery displayed 14 top abnormal cells in viewing software for expert analysis
- 1 cytopathologist (1 CP) and 2 cytologists (CP) assigned a TPS 2.0 diagnosis to each case: NHGUC, AUC, SHGUC or HGUC; results were compared between reviewers
RESULTS:
- CP demonstrated the highest specificity, PPV, accuracy and agreement with the ground truth compared with the cytologists (CPs), who had greater sensitivity and NPV
- Both CP and CTs had reduced sensitivity, PPV and agreement in the diagnosis of AUC vs NHGUC
CONCLUSIONS:
Decreased sensitivity and PPV in differentiating AUC from NHGUC may be the result of the usual clinical workflow, where cytologists initially screen and mark suspicious regions as abnormal, but defer to CP for a more specific interpretation; it may also reflect less experience with digital cytology interpretation.
November 2023
SOFTWARE OPTIMIZATION & VALIDATION STUDIES
ASC Poster Presentation
Evaluating image quality and reproducibility of digital scanners for whole-slide imaging in urine cytology slides for artificial intelligence applications
Evaluating image quality and reproducibility of digital scanners for whole-slide imaging in urine cytology slides for artificial intelligence applications
ASC 2023 Poster Presentation / Abstract (JASC 2023; Vol12(5):S74-75). Yang WL, Hang JF, Ou YC et al.
STUDY DESIGN:
- 20 urine cytology slides, 5 types of preparation (cytospin, ThinPrep nonGYN, ThinPrep Urocyte, and BD CytoRich) were digitized using 3 digital scanners (Roch DP200, Roch DP600, and Hamamatsu Nanozoomer 360)
- Images were evaluated for quality, focus and color from the default, manual and advanced scanning modes by a senior cytologist
RESULTS:
- Roche DP200 and DP600 scanners achieved good quality WSI (30-50% in default mode and 40-65% in manual mode)
- Hamamatsu achieved 90% good quality imaging only in the manual mode, whereas the default quality dropped to 15%. It also had a lower coefficient of variation for average number of atypical cells across all preparation types
CONCLUSIONS:
Overall, Hamamatsu scanner showed better performance for image quality in the manual mode than Roche scanners, which performed better in the default mode. Hamamatsu scanners also had improved detection of atypical urothelial cells among triplicate images from the same samples compared to Roche scanners.
October 2023
AI SOFTWARE DEVELOPMENT
ASC Poster Presentation
Assessing an artificial intelligence-driven digital cytology tool for efficient diagnosis of upper tract urothelial carcinoma in upper urinary tract cytology
Assessing an artificial intelligence-driven digital cytology tool for efficient diagnosis of upper tract urothelial carcinoma in upper urinary tract cytology
ASC 2023 Poster Presentation / Abstract (JASC 2023; Vol 12(5):S73-75) Wei-Lei Yang; JF Hang et al.
STUDY DESIGN:
- 60 paired cytospin and CytoRich slides from 30 pts scanned into WSI and inferred using AIxURO AI-driven software, ranking into most top 24 suspicious cells/groups
- 3 senior cytologists with variable digital interpretive experience (A= over 1 year; B = 6 mo-1 yr; C = less than one month) evaluated interface gallery of images to render a diagnosis
RESULTS:
- Dx performance was consistent for the 2 preparation types, with cytologists A & B showing excellent performance
- All reviewers spent similar amounts of time on review per preparation type, with cytologist C taking the least amount of time.
CONCLUSIONS:
The AI-assisted tool allowed for excellent performance to aid interpretation of upper GU tract urothelial carcinoma
Upcoming Events
Upcoming event
Pathology Visions 2025
05
-
07
Oct 2025
San Diego, California
October
05
-
07
,
2025
Latest news
November 2024
American Society of Cytopathology (ASC)
Dr. Barbara Crothers is the recipient of the ASC’s highest award- 2024 Papanicolaou AwardApril 2024
