promptable-cancer-segmentation

KO | EN

gitlite — search

#python #ai #typescript #llm #golang #machine-learning #rust #docker #java #windows #cli #security #openai #open-source

promptable-cancer-segmentation

(★ 13)

This project addresses the challenge of prostate cancer segmentation in T2-weighted MRI scans by integrating a weakly-supervised and a fully-supervised classifier. The method uses classification and user-defined prompts to guide a spiral search, refining object presence predictions and assembling them into a structured pixel-level segmentation mask

promptable-cancer-segmentation 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Citation

Please cite as below:

Lynn Karam, Yipei Wang, Veeru Kasivisvanathan, Mirabela Rusu, Yipeng Hu, and Shaheer U. Saeed. Promptable Cancer Segmentation Using Minimal Expert-Curated Data. Medical Image Understanding and Analysis (MIUA), 2025.
DOI: https://doi.org/10.48550/arXiv.2505.17915

# Promptable cancer segmentation using minimal expert-curated data

Current segmentation models have shown promising results in the medical field, however prostate cancer segmentation specifically remains a challenge. This repository contains codes for a project which integrates two classifiers, one weakly-supervised and one fully-supervised classifier, for segmentating prostate cancer in T2-weighted MRI scans.

Here is a segmentation method that combines classification and user-defined prompts to localise and delineate regions of interest in volumetric images. A crop of the volumetric image is passed through a weakly-supervised classifier. This predicts the likelihood of object presence in the crop. Simultaneously, a fully-supervised classifier is used to refine classifications of presence by using an edge detection and region overlap mechanism. The crop is then guided by both classifiers in a spiral search, which allows selection of regions where the object is likely present. These regions are then assembled into a pixel-level segmentation mask by combining predictions into a structured output.

The following figure describes how the framework works.

Use the split_data.py code to split your data into training, testing and validating datasets. Then, use the latter to train the weakly-supervised object presence classifier on the weak_classifier.py code, and save the weights. Upload those weights into the click_prompt_app.py if you want to manually prompt on the image and visualise it, or use the click_prompt_auto.py if you want automatic prompting through the dataset.

promptable-cancer-segmentation

promptable-cancer-segmentation 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

promptable-cancer-segmentation

promptable-cancer-segmentation 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README