Muhammad Nurmahir Mohamad Sehmi, M. F. A. Fauzi, Wan Siti Halimatul Munirah Wan Ahmad, E. Chan
{"title":"胰:基于云的胰腺癌自动分级系统","authors":"Muhammad Nurmahir Mohamad Sehmi, M. F. A. Fauzi, Wan Siti Halimatul Munirah Wan Ahmad, E. Chan","doi":"10.3389/frsip.2022.833640","DOIUrl":null,"url":null,"abstract":"Pancreatic cancer is one of the deadliest diseases which has taken millions of lives over the past 20 years. Due to challenges in grading pancreatic cancer, this study presents an automated cloud-based system, utilizing a convolutional neural network deep learning (DL) approach to classifying four classes of pancreatic cancer grade from pathology image into Normal, Grade I, Grade II, and Grade III. This cloud-based system, named PancreaSys, takes an input of high power field images from the web user interface, slices them into smaller patches, makes predictions, and stitches back the patches before returning the final result to the pathologist. Anvil and Google Colab are used as the backbone of the system to build a web user interface for deploying the DL model in the classification of the cancer grade. This work employs the transfer learning approach on a pre-trained DenseNet201 model with data augmentation to alleviate the small dataset’s challenges. A 5-fold cross-validation (CV) was employed to ensure all samples in a dataset were used to evaluate and mitigate selection bias during splitting the dataset into 80% training and 20% validation sets. The experiments were done on three different datasets (May Grunwald-Giemsa (MGG), hematoxylin and eosin (H&E), and a mixture of both, called the Mixed dataset) to observe the model performance on two different pathology stains (MGG and H&E). Promising performances are reported in predicting the pancreatic cancer grade from pathology images, with a mean f1-score of 0.88, 0.96, and 0.89 for the MGG, H&E, and Mixed datasets, respectively. The outcome from this research is expected to serve as a prognosis system for the pathologist in providing accurate grading for pancreatic cancer in pathological images.","PeriodicalId":93557,"journal":{"name":"Frontiers in signal processing","volume":"137 1","pages":""},"PeriodicalIF":1.3000,"publicationDate":"2022-02-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"PancreaSys: An Automated Cloud-Based Pancreatic Cancer Grading System\",\"authors\":\"Muhammad Nurmahir Mohamad Sehmi, M. F. A. Fauzi, Wan Siti Halimatul Munirah Wan Ahmad, E. Chan\",\"doi\":\"10.3389/frsip.2022.833640\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Pancreatic cancer is one of the deadliest diseases which has taken millions of lives over the past 20 years. Due to challenges in grading pancreatic cancer, this study presents an automated cloud-based system, utilizing a convolutional neural network deep learning (DL) approach to classifying four classes of pancreatic cancer grade from pathology image into Normal, Grade I, Grade II, and Grade III. This cloud-based system, named PancreaSys, takes an input of high power field images from the web user interface, slices them into smaller patches, makes predictions, and stitches back the patches before returning the final result to the pathologist. Anvil and Google Colab are used as the backbone of the system to build a web user interface for deploying the DL model in the classification of the cancer grade. This work employs the transfer learning approach on a pre-trained DenseNet201 model with data augmentation to alleviate the small dataset’s challenges. A 5-fold cross-validation (CV) was employed to ensure all samples in a dataset were used to evaluate and mitigate selection bias during splitting the dataset into 80% training and 20% validation sets. The experiments were done on three different datasets (May Grunwald-Giemsa (MGG), hematoxylin and eosin (H&E), and a mixture of both, called the Mixed dataset) to observe the model performance on two different pathology stains (MGG and H&E). Promising performances are reported in predicting the pancreatic cancer grade from pathology images, with a mean f1-score of 0.88, 0.96, and 0.89 for the MGG, H&E, and Mixed datasets, respectively. The outcome from this research is expected to serve as a prognosis system for the pathologist in providing accurate grading for pancreatic cancer in pathological images.\",\"PeriodicalId\":93557,\"journal\":{\"name\":\"Frontiers in signal processing\",\"volume\":\"137 1\",\"pages\":\"\"},\"PeriodicalIF\":1.3000,\"publicationDate\":\"2022-02-11\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Frontiers in signal processing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.3389/frsip.2022.833640\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"ENGINEERING, ELECTRICAL & ELECTRONIC\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Frontiers in signal processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.3389/frsip.2022.833640","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
PancreaSys: An Automated Cloud-Based Pancreatic Cancer Grading System
Pancreatic cancer is one of the deadliest diseases which has taken millions of lives over the past 20 years. Due to challenges in grading pancreatic cancer, this study presents an automated cloud-based system, utilizing a convolutional neural network deep learning (DL) approach to classifying four classes of pancreatic cancer grade from pathology image into Normal, Grade I, Grade II, and Grade III. This cloud-based system, named PancreaSys, takes an input of high power field images from the web user interface, slices them into smaller patches, makes predictions, and stitches back the patches before returning the final result to the pathologist. Anvil and Google Colab are used as the backbone of the system to build a web user interface for deploying the DL model in the classification of the cancer grade. This work employs the transfer learning approach on a pre-trained DenseNet201 model with data augmentation to alleviate the small dataset’s challenges. A 5-fold cross-validation (CV) was employed to ensure all samples in a dataset were used to evaluate and mitigate selection bias during splitting the dataset into 80% training and 20% validation sets. The experiments were done on three different datasets (May Grunwald-Giemsa (MGG), hematoxylin and eosin (H&E), and a mixture of both, called the Mixed dataset) to observe the model performance on two different pathology stains (MGG and H&E). Promising performances are reported in predicting the pancreatic cancer grade from pathology images, with a mean f1-score of 0.88, 0.96, and 0.89 for the MGG, H&E, and Mixed datasets, respectively. The outcome from this research is expected to serve as a prognosis system for the pathologist in providing accurate grading for pancreatic cancer in pathological images.