cyyu96/AutoMSC-Baselines

AutoMSC Baseline Models

Baseline models for MICCAI FLARE 2026 Task 2 — AutoMSC: Automated Medical Image Segmentation and Classification.

Codebase: github.com/medfm-flare/AutoMSC-Baseline

Each model jointly performs organ/lesion segmentation and case-level classification using a multi-task learning nnUNet framework (nnUNetCLSTrainerMTL). Models are trained with 5-fold cross-validation; this repository contains the best-checkpoint weights (checkpoint_best.pth) for each fold.

Datasets

Dataset	Modality	Cases	Segmentation Labels	Classification Labels (cases per class)	License
Dataset001_BrainMets	T1, T1c, T2, FLAIR, CT, RTP	166	Necrotic Core (1), Enhancing Tumor (2), Edema (3)	Primary Tumor Origin: NSCLC (102), Breast carcinoma (64)	CC BY 4.0
Dataset002_MU_Glioma	T1c, T1, FLAIR, T2	591	Non-Enhancing Tumor Core (1), Surrounding Non-Enhancing FLAIR Hyperintensity (2), Enhancing Tumor (3), Resection Cavity (4)	Primary Diagnosis: GBM (446), Astrocytoma (87), Others (58)	CC-BY-NC 4.0
Dataset003_UCSD_PTGB	T1c, FLAIR, ADC	184	Tumor (1)	IDH Mutation Status: Wild-Type (169), Mutant (15)	CC-BY-NC 4.0
Dataset004_PICAI	T2, ADC, HBV	1500	Prostate (1), Tumor (2)	ISUP Grade: Benign/Indolent (847), ISUP 1 (228), ISUP 2 (234), ISUP 3 (99), ISUP 4 (40), ISUP 5 (52)	CC-BY-NC 4.0
Dataset005_LUNA25	CT	6132	Nodule (1)	Malignancy: Benign (5,589), Malignant (543)	CC BY 4.0
Dataset061_PETWB_Lung	CT, PET	490	Lung (1)	Cancer type: Other (42), Lung Cancer (448)	CC BY 4.0
Dataset062_PETWB_Liver	CT, PET	490	Liver (1)	Cancer type: Other (242), Liver Cancer (248)	CC BY 4.0

Repository Structure

DatasetXXX_<name>/
├── README.md                                    # per-fold results and figures
├── figures/                                     # box plots, ROC/PRC, confusion matrix, case samples
└── nnUNetCLSTrainerMTL__nnUNetPlans__3d_fullres/
    ├── dataset.json
    ├── dataset_fingerprint.json
    ├── plans.json
    ├── fold_0/
    │   └── checkpoint_best.pth
    ├── ...
    └── fold_4/
        └── checkpoint_best.pth