Synthetic Dataset Generation From Histopathology Images For Quantizing Necrosis In Post Neo-Adjuvant Chemotherapy Resection Specimen

• Published in: 2023 2nd International Conference on Computational Systems and Communication (ICCSC) pp. 1 - 5
• Main Authors: Saleena, T.S, P, Muhamed Ilyas, Kutty, Sajna V.M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.03.2023
• Subjects: data augmentation; patchification; Percentage of necrosis; prognostic factor; synthetic data generation
• DOI: 10.1109/ICCSC56913.2023.10143027

The death of tumor tissues after chemotherapy is the most important decision factor in the treatment plan in the case of Osteosarcoma and Renal Cell Carcinoma. The percentage of necrosis accumulated due to such chemotherapy will help the doctors to assess the severity of cancer in the patient. Deep learning algorithms can automatically segment the region of necrosis and find the volume of the region using digital histopathology images. This can reduce the inter-observer disagreement that arises at the time of manual segmentation by the pathologists. But the undersupply of data, especially in pathology images is an ever-time obstacle in deep learning algorithms. In our study we have created a synthetic dataset from 47 images which are captured and manually annotated by experienced pathologists. Each image captured is of dimensions 2592×1932, which are split into patches of size 512×512. So we can have 15 patches from each image and thereby we created a dataset of 705 samples. Different augmentation techniques can be applied to this dataset that can again increment the number.