已入深夜,您辛苦了!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您度过漫漫科研夜!祝你早点完成任务,早点休息,好梦!

Multimodal pathomics feature integration for enhanced predictive performance in gastric cancer pathology through radiomic and CNN-derived features

化学 特征(语言学) 癌症 模式识别(心理学) 人工智能 内科学 计算机科学 语言学 医学 哲学
作者
Lan Yan,Peng Zhao,Kangpeng Yan,Lixiang Peng
出处
期刊:Arabian Journal of Chemistry [Elsevier BV]
卷期号:18: 1852024-1852024
标识
DOI:10.25259/ajc_185_2024
摘要

This study aims to improve the accuracy and reliability of gastric cancer grading by creating a computational framework that combines radiomic features and deep learning data from pathology images. By merging traditional and modern modeling techniques, we seek to overcome current diagnostic challenges and build a model that can be used effectively in clinical settings. The dataset included 798 whole-slide images (WSIs) of gastric cancer, divided into over 278,000 smaller image patches categorized into four grades. Radiomic features were collected using the HistomicsTK tool to ensure standard and consistent data collection. At the same time, deep learning features were extracted from fine-tuned CNN models (Xception, InceptionV3, DenseNet169, and EfficientNet) designed for image classification. Advanced methods like LASSO, ANOVA, mutual information (MI), and recursive feature elimination (RFE) were used to pick the most useful features. Different machine learning models, such as XGBoost, LightGBM, CatBoost, Random Forest, Support Vector Machine (SVM), and multi-layer perceptron (MLP), were trained and tested using a five-fold cross-validation process. Performance was assessed using metrics like AUC, accuracy (ACC), and F1-score, with hyperparameters fine-tuned through grid search for the best results. In the analysis using only radiomic features, XGBoost and CatBoost showed the best results, especially with RFE feature selection, achieving test AUCs of 91.1% and 91.2%, respectively, with F1-scores above 90%. When radiomic features were combined with deep learning features from all CNN models, the performance improved significantly. CatBoost with ANOVA reached a training AUC of 97.73% and a test AUC of 95.26%, while XGBoost with RFE achieved a test AUC of 96.9%. The top selected features, which included morphometric, gradient, intensity-based, and Haralick descriptors, were confirmed for their importance through q-value analysis. The combined model showed excellent general performance, with a test AUC of 94.22%, ACC of 95.80%, and an F1-score of 93.10%, proving the strength of using combined multimodal features. This study shows the advantages of combining radiomic and deep learning features for better grading of gastric cancer. In the future, this framework could be expanded to other types of cancer and integrated into clinical workflows, potentially reducing diagnostic errors and improving patient outcomes.
最长约 10秒,即可获得该文献文件

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
kang完成签到,获得积分10
刚刚
刚刚
鲜艳的谷丝完成签到 ,获得积分10
1秒前
2秒前
打打的应助被唐唐采纳,获得10
3秒前
肥皂剧发布了新的文献求助10
3秒前
4秒前
载酒裁决的家产完成签到 ,获得积分10
4秒前
chy发布了新的文献求助10
5秒前
retro完成签到 ,获得积分10
6秒前
sun完成签到,获得积分10
7秒前
8秒前
廖小明发布了新的文献求助10
9秒前
吃蛋挞了吗完成签到,获得积分10
10秒前
小蘑菇的应助被腼腆的立诚采纳,获得10
10秒前
MOJITO完成签到,获得积分10
12秒前
BSDL发布了新的文献求助10
14秒前
舒化奶完成签到,获得积分10
14秒前
聪明的小乌龟完成签到,获得积分10
14秒前
州zzzz完成签到 ,获得积分10
15秒前
前方完成签到 ,获得积分10
17秒前
夜寻完成签到 ,获得积分10
18秒前
ket的应助被晓鸭的平凡世界采纳,获得10
18秒前
汉堡包的应助被唐唐采纳,获得10
19秒前
美女娜完成签到 ,获得积分10
20秒前
Luo完成签到 ,获得积分10
21秒前
科研通AI2S的应助被科研通管家采纳,获得10
21秒前
Nole的应助被科研通管家采纳,获得30
21秒前
null的应助被科研通管家采纳,获得10
21秒前
21秒前
orixero的应助被科研通管家采纳,获得10
21秒前
大个的应助被科研通管家采纳,获得10
22秒前
天天快乐的应助被科研通管家采纳,获得10
22秒前
希望天下0贩的0的应助被chy采纳,获得10
22秒前
better完成签到 ,获得积分10
22秒前
OK的应助被初景采纳,获得100
22秒前
23秒前
gsgtshy完成签到 ,获得积分10
23秒前
水晶鞋完成签到 ,获得积分10
25秒前
25秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
八旗制度の研究 600
高分子基功能界面水蒸发材料与应用研究 500
外场增强的水汽发电研究 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8184723
求助须知:如何正确求助?哪些是违规求助? 9573702
关于积分的说明 21832740
捐赠科研通 7782090
什么是DOI,文献DOI怎么找? 3434404
关于科研通互助平台的介绍 2558528
邀请新用户注册赠送积分活动 2481054