ICACT20240345 Slide.15
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Thank You for Your Attention
|
|
ICACT20240345 Slide.14
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Future Work.
Future work could explore additional enhancements to the proposed method, potentially incorporating region-level segmentation for comparison with fully supervised learning.
This approach opens avenues for improving diagnostic accuracy and supporting pathologists in their analysis.
|
|
ICACT20240345 Slide.13
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Conclusion
This research proposes an efficient method for detecting and classifying gastric cancer using Vision Transformers and weakly annotated histopathological images.
Unlike CNN approaches, the ViT-based model leverages global features extracted from region-level annotations, minimizing information loss.
The results show the model's potential to assist pathologists in decision-making.
|
|
ICACT20240345 Slide.12
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Results and Discussion
|
|
ICACT20240345 Slide.11
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Results and Discussion
|
|
ICACT20240345 Slide.10
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Figure 3: Architecture of Vision transformer.
|
|
ICACT20240345 Slide.09
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Vision Transformer Architecture
Vision Transformers (ViTs) are deep learning models designed for computer vision tasks. The model architecture includes a self-attention mechanism that captures relationships between different parts of an image.
The ViT model used a 16x16 patch size, generating 16,384 patches for a single input image of size 2048x2048.
The patch images were tokenized and flattened, and a linear projection layer transformed them into lower-dimensional vectors for further processing.
|
|
ICACT20240345 Slide.08
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Figure 2: Preprocessed tissue samples. (a) Weakly annotated tissue sample. (b) Extracted region of interest from (a) by removing the background. (c) Example sample of normal tissue. (d) Example samples of cancer tissue.
|
|
ICACT20240345 Slide.07
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Methods
weakly annotated regions of whole slide images (WSIs) were used for detecting and classifying gastric cancer.
The process involved manual weak annotation for weakly supervised learning. Tissue samples of size 2048x2048 were generated from these weakly annotated regions, enhancing scalability and enabling precise analysis.
The data preprocessing steps involved operations on patch images to prepare them for analysis.
|
|
ICACT20240345 Slide.06
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Figure 1: Sample images of the dataset. (a, b) Cancer samples with annotated regions. (c, d) Normal samples.
|
|
ICACT20240345 Slide.05
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Dataset.
The histopathological samples used in this study were obtained from the GasHisSDB dataset, publicly available for research.
The dataset comprises 560 cancerous images and 140 normal images, each with a resolution of 2048x2048 pixels. The images were prepared through H&E staining at 20X magnification.
The dataset provides well-annotated images by medical experts, allowing for effective training and evaluation.
|
|
ICACT20240345 Slide.04
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Objective
The objective of this research is to detect and classify gastric cancer using Vision Transformers (ViTs) applied to weakly annotated histopathological images.
The focus is on utilizing region-level annotations to classify whole slide images (WSIs) into cancerous and normal classes. The proposed approach aims to facilitate diagnosis and support decision-making for pathologists.
|
|
ICACT20240345 Slide.03
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Vision Transformers in Medical Imaging
Vision Transformers (ViTs) have emerged as powerful tools in computer vision tasks. Originally designed for natural language processing, ViTs have been extended to image processing tasks.
In medical imaging, ViTs have demonstrated efficacy in tasks such as image classification, object detection, and segmentation. Several studies, including prostate cancer detection and Covid-19 diagnosis, have showcased the effectiveness of ViTs in medical applications.
|
|
ICACT20240345 Slide.02
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Introduction.
Gastric Cancer (GC) is a prevalent and serious global health concern, ranking as the fifth most diagnosed cancer worldwide. Timely diagnosis is crucial for better outcomes.
Biopsy, a common procedure, aids in detecting abnormal areas in the stomach. The study focuses on leveraging Vision Transformers (ViTs) to detect and classify gastric cancer from weakly annotated histopathological images.
|
|
ICACT20240345 Slide.01
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Index
Introduction
Vision Transformers in Medical Imaging
Objective
Dataset
Methods
Vision Transformer Architecture
Results and Discussion
Conclusion
|
|
ICACT20240345 Slide.00
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
.jpg) |
Vision Transformer-based Model for Gastric Cancer Detection and Classification using Weakly Annotated Histopathological Images.
Tagne Poupi Theodore Armand Institute of Digital Anti-Aging Healthcare Inje University Republic of Korea Supervisor: Professor Hee-Cheol Kim
|
