SAttisUNet: UNet-like Swin Transformer with Attentive Skip Connections for Enhanced Medical Image Segmentation
Bibliographic record
Abstract
Despite the numerous advancements in Convolutional Neural Networks (CNNs) and Transformers, especially in the field of medical image segmentation, two fundamental issues remain. First, the image segmentation task often struggles with effectively modelling global contexts with multi-scales to achieve accurate segmentation results. The second issue concerns the computational burden associated with processing high-resolution medical images and producing fine-grained predictions. Dealing with this level of detail, demands significant computational resources, leading to a computationally intensive process. UNet-like encoder-decoder architectures, which are still the number one widely used architecture in many state-of-the-art applications, struggle to address these complications. While UNet's naive skip connections help to recover spatial information, they fall short in capturing the hierarchical relationships at different scales and the overall context of the image as they combine features from different layers without accounting for their differences, which leads to less accurate segmentation results. We propose an enhanced UNet-like Transformer-based framework with attentive skip connections to tackle these problems: first, instead of simply integrating features extracted from the encoder with the decoder, we added a Transformer-based skip connection module, and second, we optimized the calculations within the skip connection module by employing a merging cross-covariance attention mechanism rather than the conventional self-attention operation, which not only bridges the gaps between multiple levels of semantics and captures more complex dependencies but can also process high-resolution images more efficiently due to its linear complexity in the number of tokens. While retaining the U-shaped encoder-decoder structure, we also replace UNet's CNN layers with hierarchically equivalent Swin Transformer blocks, capturing both global interactions and local dependencies.
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 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 | 0.001 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, 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".