📄Работа №186998
Тема: TUMOR CELLS DETECTION USING MACHINE LEARNING FROM IMAGES OF ADENOCARCENOMA TISSUES
Тип работы
Бакалаврская работа
Предмет
биология
Объем:
59 листов
Год:
2025
Просмотров:
48
4500
руб.
🛒 Купить работу
Не подходит эта работа?
Закажите новую по вашим требованиям
Узнать цену на написание
Закажите новую по вашим требованиям
Представленный материал является образцом учебного исследования, примером структуры и содержания учебного исследования по заявленной теме. Размещён исключительно в информационных и ознакомительных целях.
Workspay.ru оказывает информационные услуги по сбору, обработке и структурированию материалов в соответствии с требованиями заказчика.
Размещение материала не означает публикацию произведения впервые и не предполагает передачу исключительных авторских прав третьим лицам.
Материал не предназначен для дословной сдачи в образовательные организации и требует самостоятельной переработки с соблюдением законодательства Российской Федерации об авторском праве и принципов академической добросовестности.
Авторские права на исходные материалы принадлежат их законным правообладателям. В случае возникновения вопросов, связанных с размещённым материалом, просим направить обращение через форму обратной связи.
Настоящий учебно-методический информационный материал размещён в ознакомительных и исследовательских целях и представляет собой пример учебного исследования. Не является готовым научным трудом и требует самостоятельной переработки.
📋 Содержание
ABSTRACT
INTRODUCTION 3
1. CLINICAL ASPECT 4
1.1 Colorectal adenocarcinoma 4
1.2 Universal tools for medical imaging 6
1.3 Convolutional neural networks in medical imaging 8
1.4 Predicting gene expression from histoscans 12
1.5 Difficulties in analyzing whole-slide images 15
2. FEATURE EXTRACTION USING UNSUPERVISED LEARNING 19
2.1 Unsupervised Machine Learning: Python implementation 19
2.2 ResNet50 Architecture 24
2.3 Processing whole-slide images 27
RESULTS 32
CONCLUSION 35
REFERENCES 36
INTRODUCTION 3
1. CLINICAL ASPECT 4
1.1 Colorectal adenocarcinoma 4
1.2 Universal tools for medical imaging 6
1.3 Convolutional neural networks in medical imaging 8
1.4 Predicting gene expression from histoscans 12
1.5 Difficulties in analyzing whole-slide images 15
2. FEATURE EXTRACTION USING UNSUPERVISED LEARNING 19
2.1 Unsupervised Machine Learning: Python implementation 19
2.2 ResNet50 Architecture 24
2.3 Processing whole-slide images 27
RESULTS 32
CONCLUSION 35
REFERENCES 36
📖 Введение
Despite the progress of molecular oncology, histological analysis remains a key tool for verifying the diagnosis of colorectal adenocarcinomas (CRA). However, traditional microscopy faces fundamental limitations: high subjectivity of interpretation, labor-intensive processing of whole-slide images (WSI), and difficulty in quantifying heterogeneous tissue patterns. Digital pathology, combined with deep learning techniques, offers new ways to solve these problems by automating and standardizing analysis. Of particular interest is the use of convolutional neural networks (CNNs), which can detect subtle correlations between histological architectonics and molecular profiles of tumors. The ResNet50 architecture demonstrates unique advantages for WSI processing thanks to a residual block mechanism that prevents feature degradation in deep layers of the network.
Aim
To develop and validate a ResNet50 approach for uncontrolled detection of cancer cells and gene expression on colorectal adenocarcinoma histoscans.
Research objectives
1. Optimize the processing of whole-slide images using an adapted tile splitting strategy that minimizes context losses.
2. Implement clustering of histological structures using contrastive learning methods (SwAV).
3. To assess the correlation of the identified morphological clusters with the expression of key genes (KRAS, TP53, APC).
4. Analyze the computational efficiency of the method and the clinical interpretability of the results.
Aim
To develop and validate a ResNet50 approach for uncontrolled detection of cancer cells and gene expression on colorectal adenocarcinoma histoscans.
Research objectives
1. Optimize the processing of whole-slide images using an adapted tile splitting strategy that minimizes context losses.
2. Implement clustering of histological structures using contrastive learning methods (SwAV).
3. To assess the correlation of the identified morphological clusters with the expression of key genes (KRAS, TP53, APC).
4. Analyze the computational efficiency of the method and the clinical interpretability of the results.
✅ Заключение
The study achieved its goal, confirming the effectiveness of the adapted ResNet50 architecture for uncontrolled analysis of histological sections of colorectal adenocarcinoma. The solution of research tasks is implemented as follows:
Optimization of WSI processing is provided by a patch-oriented strategy with a tile size of 512x512 pixels and an overlap of 6.25%, which preserves contextual information at the boundaries of areas. Integration of the MobileNetV3- based CNN filter made it possible to exclude 70.6% of irrelevant fragments (artifacts, background), reducing the amount of data processed while maintaining diagnostically significant structures.
Clustering of tissue patterns was successfully implemented by the SwAV algorithm, which identified four clinically significant clusters: stroma, adenocarcinoma glands, lymphoid infiltrates, and areas associated with KRAS/TP53 expression. Validation using the Adjusted Rand Index (0.931) metric confirmed that the clusters conform to the expert markup.
Computational efficiency is achieved through parallel processing on the GPU and quantization of model weights (float16), which reduces the analysis time of 1.4 million tiles to 6.3 computational hours — 7.6 times faster than the sequential approach. Clinical interpretability is provided by visualization of t-SNE, which linksen latent representations of the network in the feature space with biologically significant structures.
Future research prospects include overcoming the domain shift through federated learning, developing interpreted architectures for clinical implementation, and creating end-to-end pipelines for multimodal integration of histological, genomic, and clinical data.
Optimization of WSI processing is provided by a patch-oriented strategy with a tile size of 512x512 pixels and an overlap of 6.25%, which preserves contextual information at the boundaries of areas. Integration of the MobileNetV3- based CNN filter made it possible to exclude 70.6% of irrelevant fragments (artifacts, background), reducing the amount of data processed while maintaining diagnostically significant structures.
Clustering of tissue patterns was successfully implemented by the SwAV algorithm, which identified four clinically significant clusters: stroma, adenocarcinoma glands, lymphoid infiltrates, and areas associated with KRAS/TP53 expression. Validation using the Adjusted Rand Index (0.931) metric confirmed that the clusters conform to the expert markup.
Computational efficiency is achieved through parallel processing on the GPU and quantization of model weights (float16), which reduces the analysis time of 1.4 million tiles to 6.3 computational hours — 7.6 times faster than the sequential approach. Clinical interpretability is provided by visualization of t-SNE, which linksen latent representations of the network in the feature space with biologically significant structures.
Future research prospects include overcoming the domain shift through federated learning, developing interpreted architectures for clinical implementation, and creating end-to-end pipelines for multimodal integration of histological, genomic, and clinical data.
ИЛИ
Поиск аналога
📕 Список литературы
🖼 Скриншоты
🛒 Оформить заказ
⚡ Работу высылаем в течении 5 минут после оплаты.