Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma
In digital pathology, deep learning has been shown to have a wide range of applications, from cancer grading to segmenting structures like glomeruli. One of the main hurdles for digital pathology to be truly effective is the size of the dataset needed for generalization to address the spectrum of po...
| Main Authors: | , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-01-01
|
| Series: | Journal of Pathology Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2153353922007404 |
| _version_ | 1828086689416871936 |
|---|---|
| author | Jonathan Folmsbee Lei Zhang Xulei Lu Jawaria Rahman John Gentry Brendan Conn Marilena Vered Paromita Roy Ruta Gupta Diana Lin Shabnam Samankan Pooja Dhorajiva Anu Peter Minhua Wang Anna Israel Margaret Brandwein-Weber Scott Doyle |
| author_facet | Jonathan Folmsbee Lei Zhang Xulei Lu Jawaria Rahman John Gentry Brendan Conn Marilena Vered Paromita Roy Ruta Gupta Diana Lin Shabnam Samankan Pooja Dhorajiva Anu Peter Minhua Wang Anna Israel Margaret Brandwein-Weber Scott Doyle |
| author_sort | Jonathan Folmsbee |
| collection | DOAJ |
| description | In digital pathology, deep learning has been shown to have a wide range of applications, from cancer grading to segmenting structures like glomeruli. One of the main hurdles for digital pathology to be truly effective is the size of the dataset needed for generalization to address the spectrum of possible morphologies. Small datasets limit classifiers’ ability to generalize. Yet, when we move to larger datasets of whole slide images (WSIs) of tissue, these datasets may cause network bottlenecks as each WSI at its original magnification can be upwards of 100 000 by 100 000 pixels, and over a gigabyte in file size. Compounding this problem, high quality pathologist annotations are difficult to obtain, as the volume of necessary annotations to create a classifier that can generalize would be extremely costly in terms of pathologist-hours. In this work, we use Active Learning (AL), a process for iterative interactive training, to create a modified U-net classifier on the region of interest (ROI) scale. We then compare this to Random Learning (RL), where images for addition to the dataset for retraining are randomly selected. Our hypothesis is that AL shows benefits for generating segmentation results versus randomly selecting images to annotate. We show that after 3 iterations, that AL, with an average Dice coefficient of 0.461, outperforms RL, with an average Dice Coefficient of 0.375, by 0.086. |
| first_indexed | 2024-04-11T04:59:48Z |
| format | Article |
| id | doaj.art-5459228b1f3e4a84862e8f5585070ad6 |
| institution | Directory Open Access Journal |
| issn | 2153-3539 |
| language | English |
| last_indexed | 2024-04-11T04:59:48Z |
| publishDate | 2022-01-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Journal of Pathology Informatics |
| spelling | doaj.art-5459228b1f3e4a84862e8f5585070ad62022-12-26T04:09:01ZengElsevierJournal of Pathology Informatics2153-35392022-01-0113100146Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinomaJonathan Folmsbee0Lei Zhang1Xulei Lu2Jawaria Rahman3John Gentry4Brendan Conn5Marilena Vered6Paromita Roy7Ruta Gupta8Diana Lin9Shabnam Samankan10Pooja Dhorajiva11Anu Peter12Minhua Wang13Anna Israel14Margaret Brandwein-Weber15Scott Doyle16Department of Pathology & Anatomical Sciences, University at Buffalo SUNY, Buffalo, NY, USA; Department of Biomedical Engineering, University at Buffalo SUNY, Buffalo, NY, USA; Corresponding author at: Jacobs School 955 Main Street, Room 4205, Pathology and Anatomical Sciences, Buffalo, NY 14203, USADepartment of Pathology & Anatomical Sciences, University at Buffalo SUNY, Buffalo, NY, USAIcahn School of Medicine, The Mount Sinai Hospital, New York, NY, USADepartment of Pathology, Case Western University, Cleveland, OH, USADepartment of Pathology, Nebraska Medical Health System, Omaha, NE, USADepartment of Pathology, University of Edinburgh, Edinburgh, UKDepartment of Oral Pathology, Oral Medicine and Maxillofacial Imaging, School of Dental Medicine, Tel Aviv University, Tel Aviv, IL, USA; Institute of Pathology, Sheba Medical Center, Tel Hashomer, Ramat Gan, IL, USADepartment of Pathology, Tata Memorial Cancer Center, Mumbai, IN, USADepartment of Tissue Pathology and Diagnostic Oncology, NSW Health Pathology, Royal Prince Alfred Hospital and University of Sydney, Sydney, AU, USADepartment of Pathology, The University of Alabama at Birmingham, Birmingham, AL, USADepartment of Pathology, George Washington University Hospital, Washington, DC, USADepartment of Oncologic Surgical Pathology, Memorial Sloan Kettering Cancer Center, New York, NY, USADepartment of Pathology, University of Pennsylvania, Philadelphia, PA, USADepartment of Pathology, Yale University School of Medicine, New Haven, CT, USADepartment of Anatomic Pathology, Robert J. Tomsich Pathology and Laboratory Medicine Institute, Cleveland Clinic, Cleveland, OH, USAIcahn School of Medicine, The Mount Sinai Hospital, New York, NY, USADepartment of Pathology & Anatomical Sciences, University at Buffalo SUNY, Buffalo, NY, USA; Department of Biomedical Engineering, University at Buffalo SUNY, Buffalo, NY, USAIn digital pathology, deep learning has been shown to have a wide range of applications, from cancer grading to segmenting structures like glomeruli. One of the main hurdles for digital pathology to be truly effective is the size of the dataset needed for generalization to address the spectrum of possible morphologies. Small datasets limit classifiers’ ability to generalize. Yet, when we move to larger datasets of whole slide images (WSIs) of tissue, these datasets may cause network bottlenecks as each WSI at its original magnification can be upwards of 100 000 by 100 000 pixels, and over a gigabyte in file size. Compounding this problem, high quality pathologist annotations are difficult to obtain, as the volume of necessary annotations to create a classifier that can generalize would be extremely costly in terms of pathologist-hours. In this work, we use Active Learning (AL), a process for iterative interactive training, to create a modified U-net classifier on the region of interest (ROI) scale. We then compare this to Random Learning (RL), where images for addition to the dataset for retraining are randomly selected. Our hypothesis is that AL shows benefits for generating segmentation results versus randomly selecting images to annotate. We show that after 3 iterations, that AL, with an average Dice coefficient of 0.461, outperforms RL, with an average Dice Coefficient of 0.375, by 0.086.http://www.sciencedirect.com/science/article/pii/S2153353922007404Active learningOral cavity cancerComputational pathologyU-netDigital pathologySemantic segmentation |
| spellingShingle | Jonathan Folmsbee Lei Zhang Xulei Lu Jawaria Rahman John Gentry Brendan Conn Marilena Vered Paromita Roy Ruta Gupta Diana Lin Shabnam Samankan Pooja Dhorajiva Anu Peter Minhua Wang Anna Israel Margaret Brandwein-Weber Scott Doyle Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma Journal of Pathology Informatics Active learning Oral cavity cancer Computational pathology U-net Digital pathology Semantic segmentation |
| title | Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| title_full | Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| title_fullStr | Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| title_full_unstemmed | Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| title_short | Histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| title_sort | histology segmentation using active learning on regions of interest in oral cavity squamous cell carcinoma |
| topic | Active learning Oral cavity cancer Computational pathology U-net Digital pathology Semantic segmentation |
| url | http://www.sciencedirect.com/science/article/pii/S2153353922007404 |
| work_keys_str_mv | AT jonathanfolmsbee histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT leizhang histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT xuleilu histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT jawariarahman histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT johngentry histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT brendanconn histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT marilenavered histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT paromitaroy histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT rutagupta histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT dianalin histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT shabnamsamankan histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT poojadhorajiva histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT anupeter histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT minhuawang histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT annaisrael histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT margaretbrandweinweber histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma AT scottdoyle histologysegmentationusingactivelearningonregionsofinterestinoralcavitysquamouscellcarcinoma |