Maximilian Dressler, M.D. · University of Heidelberg
| Aim | Research Question | Proof-of-Concept Project | Repository | Data Status |
|---|---|---|---|---|
| Aim 1 | Transparent station-level prevalence visualization for professor-facing review | Interactive descriptive prototype for 11 perigastric stations using published T-stage prevalence; individualized calibration remains future work | the-station-risk-map | ⚠ Scaffold |
| Aim 2 | Quality metric extraction linked to survival outcomes | TNM heatmaps, Kaplan-Meier curves, stage distributions from TCGA STAD | the-gastric-cancer-staging-visualization | ✓ Public data |
| Aims 1–2 | Ensuring clinical trustworthiness: discrimination, calibration, clinical utility | TRIPOD/PROBAST validation implementation (AUC, Brier, ECE, decision curves) | the-medical-ai-validation | ✓ TCGA STAD |
| Foundation | Surgical instrument segmentation from laparoscopic video | DeepLabV3 pipeline: CholecSeg8k CV plus SISVSE transfer run; zero-shot IoU 0.064 → fine-tuned IoU 0.491 on 480 public robotic-gastrectomy frames | the-surgical-instrument-segmentation | Public transfer evidence |
All repositories: github.com/Herbert-Research · Independent building blocks on public data — not an integrated system. Scaffold values are literature-derived placeholders, not validated for clinical use.
| RQ1 | Can preoperative variables estimate station-specific nodal risk for KLASS-standardized surgical planning? |
| RQ2 | Can gastrectomy video features improve risk reasoning and surgical-quality assessment? |
| RQ3 | Can public CholecSeg8k/SISVSE transfer work become valid SNUH robotic-gastrectomy evidence after IRB/privacy approval? |
IRB approval · Data governance · Video capture protocols · Baseline station-specific risk models from KLASS registry (Aim 1) · Pre-register analyses · 60–80 cases
Intraoperative visual feature integration · Quality metric extraction + reliability validation (Aim 2) · Feasibility study · 150–200 total cases
Internal validation with subgroup analyses · Multimodal fusion pilot (Aim 3, exploratory) · Dissertation · Multicenter protocol drafted
This research requires high-volume KLASS-standardized gastrectomy, mature prospective registries, established quality metrics (KLASS-02-QC), and active surgical technology research. No other institution combines these elements. My 11-week immersion on Prof. Lee's service and our co-authored publication confirm productive collaboration. Negative results are equally valuable within this validation-first framework — demonstrating where AI does not yet warrant clinical integration provides evidence-based boundaries for deployment.