PATH-OSONG | DockMaster-SONGDO | TARGIS BUSAN
AI-Powered Pathology Platform
From a Single Slide to Expert-Level Cancer Insights
Multi-Cancer Expert AI Suite : Specialized AI for lung, breast, colon, gastric, and prostate cancers—individually trained for diagnostic accuracy.
Transparent AI Decisions : Heatmaps + logic ensure traceable, auditable cancer decisions.
HybridLLM-CNN Engine : Combines pattern recognition with medical reasoning for higher accuracy.
Regulatory-Grade & SaMD-Ready : One-click FDA/EMA/MFDS-ready XAI reports with SaMD-level traceability and auditability.
Future-Proof Architecture : Modular design supports IHC, rare cancers, and trial endpoints.
1. Lung adenocarcinoma
Nuclear pleomorphism+ invasion boundary
segmentation
Nuclear pleomorphism+ invasion boundary
segmentation
Detection of subtle alveolar wall thickening and nuclear pleomorphismis critical. AI improves reproducibility in AIS vs. invasive subtype classification. Accurate boundary identification correlates with prognostic staging.
constnuclei = detectNucleiContours(image); constfiltered = filterValidNuclei(nuclei); constinvasion = detectInvasionMargins(image); consttil= calculateTILDensity(image); constatypia = analyzeNuclearAtypia(filtered);
2. Breast cancer
Glandular formation + TIL + Nottingham
style scoring
Source: Eli and Edythe Broad Center for Regenerative Medicine and Stem Cell Research at USC
Architectural patterns and mitotic count drive Nottingham scoring. Tumor-infiltrating lymphocytes (TILs) help classify triple-negative subtypes. Deep learning enhances diagnosis even with image-only input.
constglands = detectGlandStructures(image); constatypia = analyzeNuclearMorphology(nuclei); consttil= calculateTILRatio(image); constgrade = estimateNottinghamGrade({ glands, atypia, mitoticCount}); constexplanation = LLMModel.explain({ score, features });
3. Colon cancer
Gland breakage + invasion + Haralick+
necrosis
Source: Raul S. Gonzalez, M.D.
Crypt deformation and muscularismucosa invasion are key staging factors. CNN detects gland breaks and structural irregularity. Rule-based logic improves T2/T3 stage detection..
constglands = detectGlandStructures(image); constbroken = glands.filter(isBrokenOrFused); constinvasion = identifyInvasiveZones(image); constnecrosis = computeNecrosisScore(image); consttil= analyzeTILDensity(image);
4. Prostate cancer
Cribriform/glomeruloid+ gland scoring +
perineural
Nuclear pleomorphism+ invasion boundary
segmentation
Cribriform pattern is a strong indicator of high-grade cancer. Gland fusion and nuclear atypia drive automated Gleason scoring. Deep segmentation enhances prediction accuracy in GS ≥8.
constglands = segmentGlands(image); constcribiform= detectCribriformFeatures(glands); constperineural= detectPerineuralInvasion(image); constatypia = analyzeNuclearMorphology(nuclei); constgrade = estimateGleasonPattern({ cribiform, atypia })
5. Gastric cancer
Signet-ring cell + Linitisplastica+ necrosis
Source: Patrick Lynch, MD, JD, University of Texas, MD Anderson Cancer Center
Signet-ring cell carcinoma is often missed by the naked eye. AI segmentation detects mucin-filled cells with displaced nuclei. Helps distinguish diffuse type from chronic gastritis.
constsignet = detectSpecialMorphologies(image, 'signetRing'); constlinitis= detectDiffuseInfiltration(image); constnecrosis = computeNecrosisScore(image); constglands = detectGlandularPatterns(image); constgrade = estimateMalignancyGrade({ signet, glands });
Technical Features Summary
•Combined traditional analysis with modular architecture
•LLM is strictly used for confidence scoring and interpretability only
•Heatmapvisualization is supported for each cancer boundary prediction
•WHO/IASLC/Nottingham standard integration with automation
•Ensemble of 4 Expert AIs for Independent Scoring and Diagnostic Consensus
•Offline mode for demo environments
•No image upload –browser-based only
•Exportable XAI Audit Logs (JSON format) for model transparency