Ethosoft
Competition projectColonoscopy · Segmentation

Colonoscopy Polyp Segmentation

A multi-dataset study of polyp masks in colonoscopy frames, built around reproducible preprocessing and comparative segmentation experiments.

Project record

This page reports the maturity and evidence documented in Ethosoft's working archive. It does not imply peer review, regulatory clearance, or clinical validation.

Methods and evidence

Competition projectColonoscopy · Segmentation

Colonoscopy Polyp Segmentation

A multi-dataset study of polyp masks in colonoscopy frames, built around reproducible preprocessing and comparative segmentation experiments.

Models explored
Attention U-Net, U-Net++, DeepLabV3+, PVT-Cascade, and SSFormer-S with a progressive refinement network.
Ablations
Learning-rate comparisons, bilateral filtering, elastic deformation, image generators, and cosine-annealing warm restarts.
Reported results
The report records U-Net++ validation mean IoU of 0.8907 and PolypDB mean IoU of 0.8655, alongside uneven results across external datasets.

Generalization note: the documented cross-dataset variance is retained because aggregate validation performance alone can hide distribution shift.

Expanded project brief

Segmentation under dataset shift.

The polyp work compares several segmentation families—Attention U-Net, U-Net++, DeepLabV3+, PVT-Cascade, and SSFormer-S with progressive refinement—on colonoscopy imagery. Its practical emphasis is the experiment loop: resizing and mask preparation, augmentation, learning-rate studies, bilateral filtering, elastic deformation, and cosine-annealing warm restarts are treated as variables rather than hidden implementation details.

The documented validation results are paired with external-dataset behavior instead of being presented as one universal score. That matters for colonoscopy because camera systems, bowel preparation, lighting, lesion size, and annotation conventions can change the visual distribution. The project record therefore keeps both the stronger U-Net++ validation result and the less uniform cross-dataset results visible.

Next evidence: a fixed patient-level protocol, external prospective data, calibration, and lesion-size-stratified error analysis would be needed before deployment claims.

Verbatim project material

Original source text

The following text is reproduced exactly from the supplied project files. It is preserved for provenance; it may be in Turkish, use Markdown notation, or contain draft statements.

teknofest2025syz/polip/aciklama.txt
Download original file 
proje pdfte detaylıca açıklanmıştır.
teknofest2025syz/polip/zFbc67p8cyrc54lNgMW2J7keU3dxlqpC.pdf
Download original file

Original binary file preserved for download.