Training-Free Medical Image Segmentation vs Trained Networks

A multi-domain data-efficiency study — frozen-foundation-model assembly (an LLM polygon drawer in the deployed path on 1/6 domains, else SAM / op_sam / SegGPT) + a numpy genome, zero gradient training on the target task.

6 modalitiesmeasured, from scratch GT-free at inferenceLLM-agnostic skills updated · 2026-07-07

Data-efficiency

The crossover — one panel per domain

Dice vs. label budget: training-free LLM vs. the strongest baseline = max(trained nets, few-shot SAM-HQ ft) per budget, from the same combined grid as the detailed CI figure below (so they agree). Error bars: LLM = test-set bootstrap 95% CI (deterministic run); baseline = 3-seed 95% t-CI (hollow □ = † fewer than 3 seeds, no CI drawn).

▬ LLM training-free · □/▪ strongest baseline (nets ∪ SAM-HQ ft) · large dot = N=100 · ▽ = baseline below the panel floor

ISIC 2018 · dermoscopyLLM leads all N

Kvasir · endoscopy polypLLM leads all N

MSD Spleen · CTLLM ends ahead

MSD Prostate · MRIbase ends ahead

Chest X-ray · lungLLM leads all N

HC18 FetalHead · ultrasoundunder revision

The trained-NN grid is now fully locked (30/30 3-seed, incl. the CXR N=100 nnU-Net). The SAM-HQ ft baseline now uses the LLM-generated box at inference (fair localization), not the deployed pipeline box — with a fair box, few-shot SAM-HQ collapses on localization-hard domains (Kvasir ~0.3, Spleen ~0.25), so the strongest baseline reverts to the trained nets there and the LLM leads Kvasir at low N. Fair SAM-HQ is the strongest baseline only at ISIC N=5 (hollow □ = no seed-level CI available). FetalHead is withheld — temporarily removed pending a subject-disjoint re-split + clean re-eval (HC18 same-pregnancy planes were split across train/test); its old numbers are not shown. Numbers are the refreshed grid; the deployed Results table above is unchanged.

Detailed data-efficiency figure — 5 settled domains, LLM vs strongest baseline (locked nets ∪ few-shot SAM-HQ ft), both curves with 95% CIs (FetalHead withheld)

Matched data-efficiency: training-free LLM pipeline (flat, measured) vs best trained NN across N — Five settled domains, five budgets (**FetalHead withheld** — leak fix, see below). Each panel: the training-free LLM (teal) vs the **strongest baseline = max(locked trained nets, few-shot SAM-HQ ft)** (blue dashed). The trained-NN grid is now **fully locked (30/30 3-seed)**, including the CXR N=100 nnU-Net. Both curves carry **95% CIs** — LLM = test-set bootstrap (deterministic run); baseline = 3-seed 95% t-CI (hollow □ = fewer than 3 seeds, no CI). Green shading marks where the LLM leads. The **SAM-HQ ft** baseline uses the **LLM-generated box at inference (fair localization)**, not the deployed pipeline box; with a fair box, few-shot SAM-HQ **collapses on localization-hard domains** (Kvasir ~0.3, Spleen ~0.25), so the strongest baseline reverts to the trained nets there and **the LLM leads Kvasir at low N**. Fair SAM-HQ is the strongest baseline only at **ISIC N=5** (hollow □, no seed-level CI). The deployed Results table above is unchanged.

Results — best trained NN across N vs the frozen-assembly pipeline

Strongest trained NN per cell — max of SegFormer (mit_b2), pretrained U-Net (resnet34), and nnU-Net, 2-seed means — at matched budgets N = 10 / 25 / 50 / 100, against the LLM pipeline (re-measured from scratch via eval_all_domains.py). The LLM's only label-fit module is a ~5–8-param genome; localization and SAM are GT-free/frozen — so the LLM number is ~flat in N while the NN climbs. Caveat: the N=100 column is a set of within-noise near-ties — best-NN@100 sits inside the LLM bootstrap 95% CI on 5/6 domains (stats.md), and 0/4 apparent N=100 leads survive as significant. The honest, CI-separated lead is at N≤25. Amber = within-CI near-tie; a green cell reflects the point estimate only, not significance.

Domain	best NN N=10	N=25	N=50	N=100	training-free LLM	few-shot decoder N=100 (oracle)
ISIC 2018 (derm)	0.815	0.829	0.832	0.868	0.880	0.911
MSD Spleen (CT)	0.633	0.802	0.859	0.907	0.920	0.953
Kvasir-SEG (endo)	0.756	0.824	0.833	0.889	0.8646	0.944
ChestXray (X-ray)	0.938	0.948	0.952	0.9546	0.9497 SegGPT in-context; fs-deploy 0.9488	0.9535
HC18 Fetal-Head (US)	— under revision (HC18 subject-level leak; withheld)
MSD Prostate (MRI) ·new	0.679	0.720	0.772	0.815	0.7877 SegGPT⊕op_sam ensemble; was 0.760	0.938

Two LLM columns: the training-free pipeline (zero gradient training — the study's headline) and the few-shot tuned decoder (generic-SAM init, encoder frozen, ~4M-param mask decoder tuned on N=100 labels, GT-box oracle). The few-shot decoder beats the from-scratch NN on four of six domains given a good box (ISIC/Spleen/Kvasir/Prostate) and ties the other two (CXR — near-ceiling NN at ~0.95; HC18 withheld) — the residual between it and the deployed training-free pipeline is mostly localization (the exception was CXR, where the frozen SAM decoder was the wall — now moved training-free by swapping in a different frozen decoder, the SegGPT in-context drawer, to 0.9497), detailed below.

CXR is now training-free again at 0.9497 — a swapped 4th frozen drawer backend (SegGPT in-context). The old CXR training-free headline (0.922, frozen box-SAM) is superseded by a different frozen decoder: SegGPT (BAAI/seggpt-vit-large, generic → leak-free, non-SAM) paints the whole lung mask in-context from K=8 spread-sampled train (image,mask) support pairs — no box, no SAM, no localizer, ZERO gradient training. Deployed 0.9497 (median 0.9616, n=107), within 0.005 of the from-scratch NN's 0.9546 — a near-tie, still not a win. This is the 4th backend at the model-coupled drawer seam (LLM polygon · frozen SAM · few-shot tuned decoder · in-context SegGPT): the seam is genuinely swappable, and CXR's gap was neither architectural nor requiring our labels — it was the particular frozen decoder (SAM). A separate, non-training-free few-shot lane remains documented: tuning only the ~4M-param SAM mask decoder on N=100 CXR labels (generic-vit_l init, encoders frozen, box-jittered, leak-free) + mask-bbox re-decode reaches 0.9488 (median 0.9608, 102/107 improve) — but the deployed CXR is now the training-free SegGPT 0.9497, which supersedes it. Frozen-decoder study (leak-free generic): among frozen alternatives the generic in-context SegGPT wins; SAM2 / SAM1-vit_h merely tie frozen SAM-HQ (no gain), and text-prompted CLIPSeg fails on grayscale X-ray (0.00) — the interesting frozen alternative is a generic in-context segmenter (few-shot at inference, zero gradient), not another SAM variant.

The pipeline wins the EXTREME scarce-label regime (N≤25) where the target is GT-free-localizable — ISIC and Spleen are CI-separated leads at N≤25 (fewshot_matched.md); at N=100 the numbers below (ISIC 0.880 vs 0.868, Spleen 0.920 vs 0.907) are within-CI near-ties, not significant leads (stats.md) — and on Kvasir polyp a frozen-DINOv2 dense-correspondence localizer (OP-SAM CPG + peak-CC box) lifts it 0.659→0.8646 — the gap was localization (frozen GT-box oracle 0.914 > NN), and this closes 89% of it; the median 0.9484 now beats the NN mean, though the mean stays just below best-NN 0.889. On Chest X-ray the deployed training-free path is now a swapped 4th frozen drawer — the SegGPT in-context drawer (K=8 support pairs, zero gradient) — reaching 0.9497 ≈ NN 0.9546 (−0.005, a near-tie), superseding the earlier box-SAM 0.922 and the separate few-shot-tuned decoder 0.9488 lane; the old "frozen-decoder cap ~0.93" was specific to SAM, not to architecture. HC18 fetal-head is withheld — under revision pending a subject-disjoint re-split (HC18 subject-level leak); its numbers are not shown as valid results. Honest headline: at N=100 these are within-CI near-ties — 0/4 apparent leads survive bootstrap 95% CIs (stats.md), and the Spleen "+0.019" is a seed-averaging artifact (best single U-Net seed 0.9323 > LLM 0.9262). The defensible lead is at N≤25 on localizable targets (ISIC / Kvasir / Spleen / Prostate), budget-matched and CI-separated. All 6 deployed paths are gradient-free frozen-model assembly, but the LLM drawer is out of the mask-producing path on 5/6 (llm_ablation.md) — Prostate, the last domain lifted off its floor, is a complementary 2nd frozen drawer (a SegGPT⊕op_sam edge-gated keep-best ensemble). SegFormer is the strongest NN in 10 of 12 sweep cells.

Segmentation galleries

Per-case GT-vs-prediction overlays for the deployed pipeline on every test image (all 465, sorted worst-Dice first) — open the 6-domain gallery hub → (ISIC · Kvasir · Spleen · Prostate · CXR · FetalHead). Green = ground truth, red = deployed prediction; overlays rendered from the committed mask_cache/.

Where is the wall — the decoder, or localization? A fair few-shot probe

Is the residual to the trained NN a decoder limit or a localization limit? MedSAM (SAM fine-tuned on ~1M medical masks) beats box-prompted SAM — but it is data-leaky (its training set likely includes these test distributions), so it is really "import a trained medical net." The fair, leak-free counterpart of the same operation: init from the generic SAM (no medical data), freeze the image + prompt encoders, and fine-tune only the ~4M-param mask decoder on our own N-label train split — GT never touches test. Given a good box (the GT-box oracle, which isolates the decoder), this few-shot-tuned decoder beats the from-scratch NN at every budget on 4 of 6 domains (ISIC/Spleen/Kvasir/Prostate — Prostate by +0.12) and ties on the other two (CXR — near-ceiling NN at ~0.95; HC18 withheld), and it saturates at N=10 (pretrained features + a tiny adapted head), so its curve is flat-high while the NN climbs:

Domain (tuned decoder, oracle)	N=10	N=25	N=50	N=100	best NN N=100
ISIC 2018 (derm)	0.905	0.904	0.907	0.911	0.868
MSD Spleen (CT)	0.957	0.960	0.960	0.953	0.907
Kvasir-SEG (endo)	0.936	0.934	0.937	0.944	0.889
ChestXray (X-ray)	0.933	0.946	0.955	0.9535	0.9546
HC18 Fetal-Head (US)	— under revision (HC18 subject-level leak; withheld)
MSD Prostate (MRI) ·new	0.906	0.931	0.932	0.938	0.815

Mostly localization — with CXR the telling exception. The oracle hands the decoder a good box; on most domains the gap between it and the deployed pipeline is localization. Kvasir is the proof: tuned-decoder oracle 0.944 vs the original deployed 0.659 — 0.285 of pure localization headroom, zero decoder; the frozen dense-correspondence localizer has since realized most of it (0.659→0.8646), exactly as this predicted, leaving only ~0.05 to the oracle box. Where localization is already solved, the tuned decoder is a clean deployable win over both production and the NN: CXR few-shot decoder 0.9488 > the earlier box-SAM production 0.922 ≈ NN 0.9546 (though the deployed training-free CXR is now the SegGPT in-context drawer 0.9497, which supersedes both); Spleen (3D volume-prior box) 0.942 > production 0.920 > NN 0.907. This is few-shot (N=100 labels = the NN's budget), not training-free — an explicit, leak-free counterpart to MedSAM. CXR is the exception that sharpens the rule: its wall was the particular frozen decoder (SAM), not localization and not architecture — swapping in a different frozen decoder, the SegGPT in-context drawer, moves it training-free to 0.9497 (≈ NN, zero gradient), and a few-shot-tuned SAM decoder independently reaches 0.9488. So of the residual "structural / architectural ceilings," the polyp boundary was a localization gap in disguise, while the CXR "decoder wall" was a SAM-specific wall breachable by another frozen decoder with none of our labels.

Is the medical pretraining (MedSAM) worth it? — a leaky benchmark

MedSAM is SAM already fine-tuned on ~1M medical masks — a leaky upper reference (its training likely saw these test distributions). Arch-matched (both vit_b, GT-box oracle, N=100), few-shot tuning of the GENERIC SAM matches or beats the leaky MedSAM on 3 of 6 domains, and stays within 0.015 on the other 3 — its own pretraining distributions (derm/US/MRI). Medical pretraining buys no consistent edge once the decoder is adapted, and where generic SAM is already strong (Kvasir/Spleen) it starts above MedSAM even frozen. Bold = better of the two few-shot columns.

Domain (N=100 oracle, vit_b)	from-scratch NN	generic frozen	generic few-shot (fair)	MedSAM frozen (leaky)	MedSAM few-shot (leaky)
ISIC 2018 (derm)	0.868	0.870	0.915	0.921	0.943
MSD Spleen (CT)	0.907	0.941	0.956	0.909	0.946
Kvasir-SEG (endo)	0.889	0.917	0.936	0.888	0.918
ChestXray (X-ray)	0.9546	0.925	0.954	0.935	0.952
HC18 Fetal-Head (US)	— under revision (HC18 subject-level leak; withheld)
MSD Prostate (MRI) ·new	0.815	0.895	0.928	0.807	0.941

Fair few-shot generic SAM vs leaky MedSAM benchmark, N=100 oracle — Fair few-shot tuning of **generic** SAM (dark blue) vs the leaky **MedSAM** (orange), each frozen + few-shot-tuned, vs the from-scratch NN (gray) — arch-matched vit_b, N=100 GT-box oracle. The fair generic few-shot ≥ leaky MedSAM few-shot on Spleen/Kvasir/ChestXray (**3/6**); MedSAM leads on its own pretraining distributions — derm (ISIC), US (HC18), MRI (prostate) — but by **<0.015** once the generic decoder is tuned, and generic-frozen still beats MedSAM-frozen on prostate (0.895 vs 0.807). Medical pretraining helps most exactly where MedSAM saw the data, yet few-shot tuning of the generic decoder nearly erases it — decoder adaptation, not medical pretraining, is the lever.

The MedSAM attack inverts once it is prompted fairly and tested OOD (ood_medsam.md). The oracle-box column above is a leaky upper reference. Give MedSAM a fair box (learned from N GT, not the oracle) and it drops +0.16..+0.59 Dice below the deployed pipeline in-distribution; take it out-of-distribution (DRIVE retinal vessels — a distribution MedSAM never trained on) and the gap widens to +0.51 while MedSAM collapses to 0.220 even handed the ORACLE box (vs ~0.90 in-distribution). MedSAM's in-distribution edge was its decoder recognizing box-shaped targets it had trained on — memorization that does not transfer; the pipeline's few-shot-refit GT-free localization is distribution-robust.

Can it be ONE domain-agnostic pipeline? — few-shot skills + a router, SAM frozen

Each domain has a distinct setup — so can it be a single domain-agnostic pipeline that few-shot-fits per domain WITHOUT tuning SAM? We test the transferable skill — a frozen-SAM-feature localizer prototype fit from N labeled images — with identical code across all six domains, SAM never touched.

Domain-agnostic few-shot skill transfer: skills few-shot-fit, topology is routed not tuned — **(A)** One fixed pipeline (prototype-sim localizer): the skill **few-shot-fits** — saturates by ~N=10 on every domain — and scores high where the target is prototype-localizable (ISIC 0.71, CXR 0.62; HC18 withheld), mid where it is a central structure (Prostate MRI 0.51), low where it needs a specialized localizer (Kvasir 0.35, Spleen 0.12). **(B)** **Routing** the per-domain topology (genome / static-box / op_sam dense-correspondence / volume-prior / ellipse-snap / volume-trajectory) + the few-shot skill recovers most of the deployed number (Kvasir 0.35→0.77→0.86, Prostate 0.51→0.76, Spleen 0.14→0.66→0.83@N=50).

Yes — a domain-agnostic pipeline = frozen SAM + a GT-free router + few-shot skills. The learning (localizer prototype + gate weights + genome) is domain-agnostic, few-shot (~10 labels; ~50 for the isodense spleen), and SAM-frozen — it transfers with identical code. The per-domain "distinct setup" you'd notice is which localizer topology to route to — a GT-free routing choice (learned config, the project's route.py), not SAM tuning. So onboarding a new domain is ~10 labels + a routing decision, zero gradient training. Robust localizer skill: the deployed default is now the op_sam frozen-DINOv2 dense-correspondence localizer (kvasir/HC18/prostate) — it transfers 10-shot support masks through the full patch cross-correlation, and superseded the earlier SAM+DINOv2 AMG-ranker ensemble on Kvasir (0.655→0.8646). The lesson holds and is now stronger: the polyp/fold and faint-skull walls yield to the right frozen feature space — dense correspondence, plus feature-diverse supports — not to distractor-verification tricks. The AMG-ranker ensemble (at/near the best single feature space on all three, Kvasir 0.65 / Spleen 0.64 / ISIC 0.67, never catastrophic where SAM-alone fails the isodense spleen 0.40) remains the LLM-agnostic fallback where a support set is too small for correspondence.

GPU-free lane — per-stage skill loop — research (default-off)

A separate lane from the deployed comparison above. This is the training-free / GPU-free pipeline — an LLM polygon draw + numpy-only skills, no SAM, no SegGPT, no gradient training. Each domain runs a per-stage loop: research → literature-grounded preprocessing → a numpy skill → a per-stage GT-free feedback gate. It is default-off research; the deployed pipeline and every Results-table / data-efficiency number above is untouched. Visual before→after recoveries: per-stage gallery →.

Domain	GPU-free base → per-stage	Δ (best N)	with-NN lane	driver stage	GT-free gate	source
CXR	0.664 → 0.854	+0.190 (deg +0.35)	null	LOCALIZE (lung atlas+CLAHE)	valid ρ+0.58	PR #78
MSD Spleen	0.555 → 0.60	+0.019..+0.043 (deg +0.247)	null	LOCALIZE (LUQ prior)	valid ρ+0.64	PR #77
HC18 FetalHead	— under revision (HC18 subject-level leak; withheld)
MSD Prostate	0.619 → 0.652	+0.033 @N=5	null	REPAIR (isodense boundary)	valid via preprocessing	PR #76
ISIC 2018	trace 0.714 → 0.923	+0.005 [CI≠0 @N=10/25/100]	null	REPAIR (color-constancy `cc_band`)	valid ρ+0.6	PR #80
Kvasir	0.347 → 0.38	headroom +0.087..+0.108, NULL	null	LOCALIZE (gate wall)	INVALID \|ρ\|<0.12	PR #79

Scorecard: 5/6 wins + 1 gate-wall null. The per-stage LOCALIZE agent wins on CXR (+0.190), Spleen (+0.043) and FetalHead (withheld — under revision, HC18 subject-level leak); the REPAIR agent wins on Prostate (+0.033) and ISIC (+0.005, CI excludes 0 @N=10/25/100, beats best-NN @N≤50; trace ISIC_0012292 0.714→0.923). Kvasir is the honest null: real oracle headroom (+0.087..+0.108) but a gate wall (every GT-free signal |ρ|<0.12) — nothing harvests it. The with-NN lane is null on all 6 (a predicted asymmetry: the deployed NN drawer already solves the stage the GPU-free lane repairs), so the NN drawer's value scales with how broken GPU-free localization is — huge on Kvasir/Spleen, ~0 on ISIC/FetalHead at N=5.

Takeaway. Where the GPU-free failure is location and a valid GT-free anatomical cue exists (CXR atlas, spleen LUQ, FetalHead ring), the per-stage LOCALIZE agent lifts the training-free lane substantially; it is an honest null where the cue is ungateable (Kvasir gate wall) or the deployed pipeline already solved location (all with-NN lanes). Honest sub-result (ISIC): a FRONT CLAHE / illumination-equalization enhancer was a negative (−0.018..−0.030, it washes out the pigment cue) and shipped gated OFF — contrast-equalization preprocessing HURTS hand-crafted-cue pipelines, the opposite of what it does for trained nets. CI completing — ISIC/Kvasir per-stage agents still finishing.

agenticSeg · training-free frozen-foundation-model assembly for medical segmentation (LLM drawer in the deployed path on 1/6 domains) · github.com/R9-Hu/agenticSeg
Skills are LLM-agnostic numpy/config; every result reproducible via eval_all_domains.py. AI-assisted (Claude Code).