TransfusionNet: Framework for cervical cancer detection using deep learning with multi-level fusion
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
This paper presents a hybrid deep learning model, TransfusionNet, comprising VGG19, ResNet50, and ViT with fusion for early detection of cervical cancer. An improved medical image preprocessing technique combining contrast-limited adaptive histogram equalization (CLAHE), Laplacian sharpening, bilateral filtering, and unsharp masking has been applied to enhance image quality. An early-layer feature fusion strategy was used to enhance feature diversity and improve decision-making, unlike conventional methods that fuse only at final layers. The fusion strategy combines features from VGG19, ResNet50, and ViT at multiple stages using concatenation and element-wise addition, allowing integration of detailed and high-level information across different levels. Two separate transition blocks were used to align VGG19 and ResNet50 outputs at different layers. TransfusionNet is trained and tested on the publicly available SIPaKMeD dataset, containing 4,049 images of five classes. TransfusionNet is also verified on Herlev and LCPSI datasets. The five-fold cross-validation was employed to mitigate overfitting and enhance models' robustness and generalizability. TransfusionNet demonstrates promising results for early cervical cancer detection, achieving state-of-the-art performance on the SIPaKMeD dataset. For batch size 128, TransfusionNet achieved an average accuracy of 99.40%, an F1-score of 99.36%, sensitivity of 99.25%, and Cohen's Kappa of 99.67% across the five folds. On the SIPaKMeD dataset, the proposed model outperformed the best-performing model by 3.77% among the cited models. On the Herlev and LCPSI datasets, TransfusionNet surpassed the top-performing methods by 6.50% and 1.22% in accuracy, respectively. Future work will focus on validating the model on larger and more diverse datasets. The source code is available at https://github.com/souravbasakshuvo/TransFusionNet . • Feature fusion strategy implemented via bi-fusion and aggregated feature fusion. • Improved preprocessing using bilateral filtering, CLAHE, and Laplacian sharpening. • Evaluation of CNNs, RNNs, and transformers for effective feature extraction. • Improved feature extraction strategy utilizing VGG19, ResNet50, and ViT.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 it