Dual Representation Learning in CNNs: Generalised Glycolytic Patterns and Organ-specific Activations
Bibliographic record
Abstract
Three-dimensional convolutional neural networks (3D CNNs) have demonstrated strong capabilities in analyzing complex textures within volumetric medical data, enabling advanced pattern recognition in oncological imaging [1,2,3,4]. This study investigates whether 3D CNNs trained for esophageal cancer detection in PET imaging develop generalizable representations of glycolytic activity. Using 486 PET/CT scans [5,6], we analyzed CNN activation patterns in primary tumors and three physiologically high-uptake organs: heart myocardium, liver, and urinary bladder. Statistical analysis revealed strong inter-organ correlations (tumor-myocardium: 0.98, tumor-liver: 0.96), yet significant differences in activation distributions between all organ pairs ($\mathrm{p}<0.05$). The urinary bladder exhibited the highest mean activation (0.181), followed by tumor (0.172), heart myocardium (0.158), and liver (0.149). Distance correlations (0.83-0.97) and mutual information (0.92-1.39) indicated substantial shared metabolic features across regions while preserving discriminative power. These results demonstrate that CNNs learn both generalized FDG uptake patterns and organ-specific signatures, capturing broad metabolic activity yet retaining the ability to distinguish pathological from physiological uptake. The strong tumormyocardium correlation suggests challenges in differentiating certain physiological uptakes from malignancy, with implications for false positive rates. Our findings provide new insights into the feature extraction mechanisms of deep learning models in oncological PET, supporting the development of more interpretable and clinically trustworthy AI tools for cancer detection.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".