MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Digital Whole Slides-Based Deep Learning for the Prediction of Treatment Outcome in Head and Neck Squamous Cell Carcinoma

H Yu1, D Jing2, W Lu1, W Lu1, J Qiu1, L Shi1*, (1) Shandong First Medical University & Shandong Academy of Medical Sciences, Taian, CN, (2) Xiangya Hospital, Central South University, Changsha, CN

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: image features of digital whole slides may provide clinical prognostic information. This study aimed to predict the treatment outcome using digital whole slides-based deep learning in head and neck squamous cell carcinoma (HNSCC).



Methods: collected digital whole slides for HNSCC patients from The Cancer Genome Atlas (TCGA) public dataset. Inclusion criteria were: (1) patients had high-quality Hematoxylin and eosin (H&E) stained FFPE slides before treatment; (2) the slides had the presence of at least one of the following HNSCC growth patterns: cancer nests, Keratin pearl and intercellular bridges; (3) the Image Type is SVS/JPEG 2; (4) patients had available treatment outcomes (complete remission (CR) or progressive disease (PD)) in follow-up. Finally, 68 patients were selected for analysis. The tumor regions on each slide were manually contoured using ASAP software by a pathologist. All slides were converted to gray images to eliminate the influence of the inconsistency in colors. Besides, all slides were rescaled to 0-255 and resized to 256*256. Image augmentation, including rotations and zooms, were applied to increase the sample size. We created a convolutional neural network (CNN) that consists of 3 convolution blocks using tensorflow version 2.0. Each convolutional block contains a Conv2D layer followed by a max pool layer. The patients were separated into the training set, test set and validation set, while the validation set was not involved in the CNN training process.



Results: epoch of the CNN was 10. The accuracy in distinguishing PD from CR of the 10th epoch was 0.84 in the test set and 0.80 in the validation set.


Conclusion: study demonstrates that digital whole slides can provide information that related to the treatment outcome in HNSCC, and deep learning can refine the prediction of HNSCC treatment outcome. This may benefit the early adjustment of treatment strategy.

Funding Support, Disclosures, and Conflict of Interest: This study was supported by the Shandong Province Key Research and Development Program (2017GSF218075) and Taishan Scholars Program of Shandong Province.

Keywords

Digital Imaging, Image Analysis, Tissue Characterization

Taxonomy

TH- Response Assessment: Radiomics/texture/feature-based response assessment

Contact Email