radiomics-ml

Name: radiomics-ml
Author: Aperivue

208stars53forks

How to install radiomics-ml?

aperivue/medsci-skills/radiomics-ml

$npx -y skills add Aperivue/medsci-skills --skill radiomics-ml

Installs into the current project.

Summary

Produce or audit a radiomics / tabular clinical-ML study — imaging or clinical features → any classical learner (penalised logistic [LASSO / ridge / elastic-net], SVM, k-NN, naive Bayes, LDA/QDA, decision tree, random forest, gradient boosting [XGBoost / LightGBM / CatBoost], sha

Files · 1

View on GitHub

SKILL.md

1	# Radiomics / Classical-ML Skill
2
3	## Purpose
4
5	Radiomics + tree-ensemble studies (features → random forest / XGBoost → a clinical outcome) are the
6	most common solo-doable clinical-ML workflow — no GPU, no engineer — and the **most commonly
7	over-optimistic**: hundreds-to-thousands of features on tens of patients, hyperparameters tuned on the
8	same folds the performance is reported from, features selected on the whole dataset, unstable features
9	never filtered, and discrimination (AUC) reported without calibration. This skill produces the pipeline
10	correctly and audits an existing one, so the clinical result survives review (Lambin 2017; CLEAR;
11	TRIPOD+AI; PROBAST-AI).
12
13	It sits beside the imaging-DL lane: where `/model-scaffold` builds a deep network, radiomics-ml
14	covers the feature-based classical-ML path. It integrates scikit-learn / xgboost / pyradiomics
15	(referenced in the emitted code); it does not reimplement them and never runs a model on real patient
16	data.
17
18	## When to use
19	- You have a radiomics or clinical/tabular feature table and want to build a random-forest / XGBoost
20	clinical prediction model that will pass statistical review.
21	- You want to audit an existing radiomics/ML pipeline for the failure modes below.
22
23	## When NOT to use
24	- Deep-learning imaging models → `/architecture-zoo` → `/model-scaffold` → `/model-validation`.
25	- Classical inferential statistics / a regression model as the estimand → `/analyze-stats`.
26	- Interpretability of a trained network → `/explainability`.
27	- Reimplementing scikit-learn / xgboost / pyradiomics → out of scope (this skill wires and audits them).
28
29	## The failure modes (what the gate enforces)
30	1. No nested CV. Tuning and reporting on the same folds inflates performance. Use nested CV or a
31	held-out test set.
32	2. High dimensionality, low events. Features ≥ events with no dimensionality reduction overfits —
33	the classic radiomics trap. Apply LASSO / PCA / a stability + redundancy filter.
34	3. Selection outside the fold. Feature selection fit on the whole dataset leaks the held-out folds.
35	Nest selection inside each training fold.
36	4. No feature stability. Radiomics features are unstable across acquisition/segmentation — filter
37	to reproducible features (ICC / test-retest).
38	5. No calibration. A clinical prediction model needs calibration (slope/intercept + a flexible
39	curve), not discrimination alone.
40	6. No external validation. A single-cohort model needs external / temporal validation for a
41	clinical claim.
42
43	## Workflow
44
45	### Phase 1 — Extract features (integrate, don't reimplement)
46	For radiomics, extract with pyradiomics under reproducible, IBSI-aligned settings (fixed bin width,
47	resampling, normalisation) — record them. For clinical/tabular data, assemble the feature table with a
48	patient/subject ID and the outcome. See `references/radiomics_ml_guide.md`.
49
50	### Phase 2 — Build the pipeline correctly
51	- Feature stability — with test-retest / multi-rater data, keep features with ICC ≥ 0.75.
52	- Nested cross-validation — outer folds estimate performance, inner folds tune; do **feature
53	selection and scaling inside each training fold** (never on the whole dataset).
54	- Dimensionality — with features ≥ events, use LASSO / a stability+redundancy filter / PCA.
55	- Model — pick from the full classical family for the t

People who liked this also viewed

6 picks

Type

radiomics-ml

208stars53forks

View on GitHub

How to install radiomics-ml?

aperivue/medsci-skills/radiomics-ml

$npx -y skills add Aperivue/medsci-skills --skill radiomics-ml

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Radiomics / Classical-ML Skill
2
3	## Purpose
4
5	Radiomics + tree-ensemble studies (features → random forest / XGBoost → a clinical outcome) are the
6	most common solo-doable clinical-ML workflow — no GPU, no engineer — and the **most commonly
7	over-optimistic**: hundreds-to-thousands of features on tens of patients, hyperparameters tuned on the
8	same folds the performance is reported from, features selected on the whole dataset, unstable features
9	never filtered, and discrimination (AUC) reported without calibration. This skill produces the pipeline
10	correctly and audits an existing one, so the clinical result survives review (Lambin 2017; CLEAR;
11	TRIPOD+AI; PROBAST-AI).
12
13	It sits beside the imaging-DL lane: where `/model-scaffold` builds a deep network, radiomics-ml
14	covers the feature-based classical-ML path. It integrates scikit-learn / xgboost / pyradiomics
15	(referenced in the emitted code); it does not reimplement them and never runs a model on real patient
16	data.
17
18	## When to use
19	- You have a radiomics or clinical/tabular feature table and want to build a random-forest / XGBoost
20	clinical prediction model that will pass statistical review.
21	- You want to audit an existing radiomics/ML pipeline for the failure modes below.
22
23	## When NOT to use
24	- Deep-learning imaging models → `/architecture-zoo` → `/model-scaffold` → `/model-validation`.
25	- Classical inferential statistics / a regression model as the estimand → `/analyze-stats`.
26	- Interpretability of a trained network → `/explainability`.
27	- Reimplementing scikit-learn / xgboost / pyradiomics → out of scope (this skill wires and audits them).
28
29	## The failure modes (what the gate enforces)
30	1. No nested CV. Tuning and reporting on the same folds inflates performance. Use nested CV or a
31	held-out test set.
32	2. High dimensionality, low events. Features ≥ events with no dimensionality reduction overfits —
33	the classic radiomics trap. Apply LASSO / PCA / a stability + redundancy filter.
34	3. Selection outside the fold. Feature selection fit on the whole dataset leaks the held-out folds.
35	Nest selection inside each training fold.
36	4. No feature stability. Radiomics features are unstable across acquisition/segmentation — filter
37	to reproducible features (ICC / test-retest).
38	5. No calibration. A clinical prediction model needs calibration (slope/intercept + a flexible
39	curve), not discrimination alone.
40	6. No external validation. A single-cohort model needs external / temporal validation for a
41	clinical claim.
42
43	## Workflow
44
45	### Phase 1 — Extract features (integrate, don't reimplement)
46	For radiomics, extract with pyradiomics under reproducible, IBSI-aligned settings (fixed bin width,
47	resampling, normalisation) — record them. For clinical/tabular data, assemble the feature table with a
48	patient/subject ID and the outcome. See `references/radiomics_ml_guide.md`.
49
50	### Phase 2 — Build the pipeline correctly
51	- Feature stability — with test-retest / multi-rater data, keep features with ICC ≥ 0.75.
52	- Nested cross-validation — outer folds estimate performance, inner folds tune; do **feature
53	selection and scaling inside each training fold** (never on the whole dataset).
54	- Dimensionality — with features ≥ events, use LASSO / a stability+redundancy filter / PCA.
55	- Model — pick from the full classical family for the t

People who liked this also viewed

6 picks

Type