BioHCL-SSSM: Complete Medical Imaging Workspace & Model Hub

Author: unknownBrain
Project: BioHCL-SSSM: Hierarchical Convolutional State-Space Hybrid Architecture for Pediatric Pneumonia Diagnosis
Dataset: Guangzhou Women and Children’s Medical Center Pediatric Chest X-Ray Cohort (Kermany et al., Cell 2018; 5,856 images, 624 official held-out test split)


πŸ“Œ Complete Repository Contents

This public repository contains the complete end-to-end workspace for the BioHCL-SSSM project, including all pre-trained checkpoints, raw experimental sweeps, modular source code, deliverable packages, evaluation figures, and peer-reviewed reference literature:

BioHCL-SSSM-Complete-Workspace/
β”‚
β”œβ”€β”€ deliverables/                              # πŸ“ OFFICIAL SUPERVISOR DELIVERABLE PACKAGES
β”‚   β”œβ”€β”€ 1_sota_benchmarks_master.csv           # 15-Model SOTA Benchmark (Literature baselines + Ours)
β”‚   β”œβ”€β”€ 2_table1_component_ablation.csv        # Table 1: Macro Component Ablation (SSSM vs Attention vs CNN)
β”‚   β”œβ”€β”€ 3_table2_layer_hyperparameter_deduction.csv # Table 2: Proposed SSSM Layer Hyperparameter Deduction
β”‚   β”œβ”€β”€ 4_table3_loss_function_weighting.csv   # Table 3: Loss Function & Class Weighting Dynamics
β”‚   β”œβ”€β”€ 5_10_fold_cross_validation_metrics.csv # Approved 10-Fold CV metrics (Folds 1 to 10 + Mean/Std)
β”‚   β”œβ”€β”€ remaining_ablations_plan.md            # Comprehensive scientific audit & execution plan
β”‚   β”œβ”€β”€ tathagat_sir_revision_report.md        # Full revision report formatted for Tathagat Sir
β”‚   β”œβ”€β”€ v2/                                    # Clean v2 supervisor release package
β”‚   β”œβ”€β”€ raw_sweeps/                            # Raw 64-run ablation search & hardware profiling logs
β”‚   └── figures/                               # Publication-ready ROC, PR, and Confusion Matrix plots
β”‚
β”œβ”€β”€ medical imaging/                           # πŸ“ FULL ORIGINAL TRAINING CODE & WEIGHTS
β”‚   β”œβ”€β”€ best_student_cv.pt                     # Best Model Checkpoint (Fold 2: 99.81% Val Acc, 0.9987 F1)
β”‚   β”œβ”€β”€ model_fold_1.pt … model_fold_10.pt     # All 10 trained cross-validation model weights (54 MB each)
β”‚   β”œβ”€β”€ ablation.csv                           # Raw 64 empirical ablation sweep runs
β”‚   β”œβ”€β”€ extracted_code.py                      # Original full pipeline code extracted from Colab
β”‚   β”œβ”€β”€ build_benchmark.py                     # Benchmark runner
β”‚   β”œβ”€β”€ build_ablations.py                     # Ablation engine runner
β”‚   β”œβ”€β”€ build_eval.py                          # Test evaluation engine
β”‚   β”œβ”€β”€ build_xai.py                           # Explainability generator (Attention, GradCAM, LIME, SHAP)
β”‚   └── *.ipynb / *.ipynb.txt                  # Full Colab training notebooks
β”‚
β”œβ”€β”€ hf_repo_staging/                           # πŸ“ CLEAN MODULARIZED SOURCE & ARTIFACTS
β”‚   β”œβ”€β”€ src/                                   # Clean modular Python package (models, dataset, loss, eval)
β”‚   β”œβ”€β”€ checkpoints/                           # Reorganized weights (cross_validation/ and ablations/)
β”‚   β”œβ”€β”€ results/                               # CSV deliverables
β”‚   └── figures/                               # Complete high-res figures & XAI heatmaps
β”‚
β”œβ”€β”€ papers to refer/                           # πŸ“ 12 PEER-REVIEWED REFERENCE PAPERS
β”‚   β”œβ”€β”€ IBMED-S-26-01139.pdf                   # Bio-HCLSSSM Reference Paper (Computers in Biology and Medicine)
β”‚   β”œβ”€β”€ CMPB-D-26-01992_R1_reviewer.pdf        # Computer Methods and Programs in Biomedicine
β”‚   β”œβ”€β”€ EAAI-26-17234_reviewer.pdf             # Engineering Applications of Artificial Intelligence
β”‚   β”œβ”€β”€ TCIV-2026-0179.pdf                     # Computer Methods in Biomechanics and Biomedical Engineering
β”‚   β”œβ”€β”€ IJIG-D-26-00102_R1_reviewer.pdf        # International Journal of Image and Graphics
β”‚   β”œβ”€β”€ JBHI-04026-2026_Proof_hi.pdf           # IEEE Journal of Biomedical and Health Informatics
β”‚   └── ... (6 additional reviewer proofs)
β”‚
β”œβ”€β”€ run_colab_remaining_ablations.py           # πŸš€ STANDALONE RUNNER FOR REMAINING EXPERIMENTS
β”‚   # Executes the 4 remaining core runs (lambda_bio in [0.2, 0.5, 1.0] and MHSA Attention baseline)
β”‚   # Takes ~10 mins on GPU and logs directly to CSV
β”‚
β”œβ”€β”€ 10_fold_metrics.csv                        # Exact 10-fold cross-validation metrics
β”œβ”€β”€ benchmark_master.csv                       # Consolidated SOTA benchmarks
β”œβ”€β”€ remaining_ablations_plan.md                # Scientific audit & plan
└── tathagat_sir_revision_report.md            # Comprehensive revision report

πŸ† Key Performance Highlights

  1. Approved 10-Fold Cross-Validation:

    • Mean Accuracy: 99.37% Β± 0.33%
    • Mean F1-Score: 0.9957 Β± 0.0022
    • Mean ROC-AUC: 0.9929 Β± 0.0039
    • Pooled Errors: Only 33 errors across 5,232 images (3,862 TP, 1,337 TN, 12 FP, 21 FN).
    • Peak Fold: Fold 2 achieved 99.81% accuracy, 0.9987 F1, 100% recall, 99.26% specificity (best_student_cv.pt).
  2. Official Held-Out Test Set (624 Images) Benchmark:

    • BioHCL-SSSM: 96.47% Accuracy, 97.22% F1, 99.37% ROC-AUC with only 14.01M parameters and 8.56 GFLOPs.
    • Outperforms standard Vision Transformers (ViT-Base: 86.57M params, 92.47% Acc) with 83.8% fewer parameters.
    • Matches/exceeds Hierarchical Shifted-Window Transformers (Swin-T: 27.52M params, 95.99% Acc) with 49.1% fewer parameters.

πŸš€ Quick Execution on New GPU (i9 + RTX 3080 Ti or Colab)

To execute the remaining 4 core ablation runs:

# 1. Clone this repository
git clone https://huggingface.co/unknownBrain/BioHCL-SSSM-Complete-Workspace
cd BioHCL-SSSM-Complete-Workspace

# 2. Install dependencies
pip install torch torchvision timm albumentations huggingface_hub scikit-learn pandas pillow

# 3. Run the automated ablation suite
python run_colab_remaining_ablations.py

πŸ“œ License & Citation

Licensed under the MIT License.
Guangzhou Women and Children’s Medical Center Chest X-Ray Cohort (Kermany et al., Cell 2018).

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Evaluation results

  • 10-Fold Cross-Validation Accuracy on Guangzhou Pediatric Chest X-Ray Cohort (Cell 2018)
    self-reported
    99.370
  • 10-Fold Cross-Validation F1 on Guangzhou Pediatric Chest X-Ray Cohort (Cell 2018)
    self-reported
    0.996
  • 10-Fold Cross-Validation AUC on Guangzhou Pediatric Chest X-Ray Cohort (Cell 2018)
    self-reported
    0.993