biswadev-9/thesis

How biswadev-9/thesis works

This is a configurable research pipeline for four-class brain MRI classification, not a deployed inference product. Its principal workflow prepares and audits a deduplicated, stratified dataset; trains image-classification branches; extracts and caches frozen features for fusion; then evaluates and explains the assembled classifier. The README also documents preprocessing and imbalance comparisons, baselines, external and degradation evaluation, ablations, and statistical reporting. Arrows follow the documented workflow; unsampled stage internals are represented by their existing source directories.

Main entry points: full_pipeline.py Image-to-logits pipeline · split_builder.py Stratified split table · bt_mri_datamodule.py MRI datamodule · ablation_study.py Ablation and comparisons

Components 26

Data preparation

Deduplication and audit src/data/components/split_builder.py
Stratified split table src/data/components/split_builder.py
Preprocessing comparison src/analysis/preprocessing_study.py
Imbalance comparison src/analysis/imbalance_study.py
Processed image cache src/prepare_dataset.py
MRI datamodule src/data/bt_mri_datamodule.py

Model experiments

Model training src/train.py
CNN and Transformer baselines src/models/components/backbones.py
Classical feature branch src/models/components/backbones.py
Multiscale quantum branch src/models/components/quantum.py
Frozen feature extraction src/extract_features.py
Split feature caches src/extract_features.py
Fusion and loss studies src/analysis/fusion_study.py
Final fusion classifier src/models/components/fusion.py
Image-to-logits pipeline src/models/full_pipeline.py

Evaluation and explanation

Internal and external evaluation src/eval.py
Degradation evaluation src/analysis/robustness.py
Explainability analysis src/analysis/explainability.py

Study analysis

Embedding analysis src/analysis/embeddings.py
Ablation and comparisons src/analysis/ablation_study.py
Research-question mapping src/analysis/rq_mapping.py
Statistical reporting src/analysis/statistical_report.py
Study reports and results src/analysis/statistical_report.py

Other components

Researcher
Raw MRI images
External MRI data

Connections 36

Researcher → Deduplication and audit: prepares study data
Raw MRI images → Deduplication and audit: scans and deduplicates
Deduplication and audit → Stratified split table: writes splits
Stratified split table → Preprocessing comparison: supplies split data
Stratified split table → Imbalance comparison: supplies split data
Stratified split table → Processed image cache: selects images
Raw MRI images → Processed image cache: filters images
Stratified split table → MRI datamodule: defines splits
Processed image cache → MRI datamodule: provides processed images
MRI datamodule → Model training: provides batches
Model training → CNN and Transformer baselines: trains
Model training → Classical feature branch: trains
Model training → Multiscale quantum branch: trains
MRI datamodule → Frozen feature extraction: provides batches
Classical feature branch → Frozen feature extraction: extracts features
Multiscale quantum branch → Frozen feature extraction: extracts features
Frozen feature extraction → Split feature caches: writes split features
Split feature caches → Fusion and loss studies: supplies cached features
Fusion and loss studies → Final fusion classifier: selects and trains
Classical feature branch → Image-to-logits pipeline: assembles branch
Multiscale quantum branch → Image-to-logits pipeline: assembles branch
Final fusion classifier → Image-to-logits pipeline: assembles classifier
MRI datamodule → Internal and external evaluation: provides internal data
External MRI data → Internal and external evaluation: provides external data
Image-to-logits pipeline → Internal and external evaluation: predicts for evaluation
Image-to-logits pipeline → Degradation evaluation: predicts degraded inputs
Image-to-logits pipeline → Explainability analysis: provides gradients and outputs
Classical feature branch → Embedding analysis: provides representations
CNN and Transformer baselines → Ablation and comparisons: supports comparisons
Internal and external evaluation → Ablation and comparisons: supplies results
Degradation evaluation → Ablation and comparisons: supplies results
Explainability analysis → Research-question mapping: supports research questions
Ablation and comparisons → Research-question mapping: maps comparisons
Ablation and comparisons → Statistical reporting: supplies predictions
Statistical reporting → Study reports and results: produces reports
Research-question mapping → Study reports and results: summarizes findings

More Jupyter Notebook diagrams: nlp-with-transformers/notebooks · datawhalechina/competition-baseline · remitchell/python-scraping · makcedward/nlpaug · bentrevett/pytorch-sentiment-analysis

Drawn by GitDiagram on September 27, 2026 from the repository's file tree, README and source files. Read as Markdown ·