Artificial intelligence can be handy in detecting lymph node metastases of common tumors

Sevim S., Bahadır M., Kartal M. S., Sak S.

https://www.ecdp2023.org/files/ecdp2023-program.pdf, Budapest, Macaristan, 14 - 17 Haziran 2023, ss.29

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Budapest
  • Basıldığı Ülke: Macaristan
  • Sayfa Sayıları: ss.29
  • Ankara Üniversitesi Adresli: Evet

Özet

A08

Artificial intelligence can be handy in detecting lymph node metastases of common tumors

Selim Sevim1 , Murat Bahadir2 , Mustafa Said Kartal3 , Serpil Dizbay Sak1 1) Pathology Department, Ankara University Medical School, Turkey 2) Software Development, Simplex Information Technologies Inc., Turkey 3) Medical School, Sivas Cumhuriyet University Medical School, Turkey Introduction

Papillary thyroid carcinoma (PTC) and breast invasive ductal carcinoma (IDC) are common carcinomas which often metastasize to regional lymph nodes (LNs). Although the accurate evaluation of LNs is very important in terms of guiding the treatment, this routine task is time-consuming and tedious. We investigated whether artificial intelligence (AI) can distinguish between metastatic and non-metastatic areas via whole slide images (WSIs) obtained from LNs of PTC and IDC. Material and methods HE slides of LNs from 44 PTC, 57 IDC (303 LNs) were scanned (3D HISTECH, Panoramic P250 Flash3 ). After manual annotation and color normalization, WSIs were used for testing (%20), training (65%), and training validation (15%). UnetPlusPlus, ImageNet, Efficientnet-b3 and Resnet were used for LN segmentation, transfer learning, feature extraction for LN detection and to detect tumors, at different zoom levels (ZL), respectively. Results and discussion AUC: 0.9717, accuracy: 0.9554, recall: 0.9702, precision: 0.8492, F-score: 0.8850 were found in LN detection/ segmentation. In the IDC and PTC groups, scores for separating tumoral and non-tumoral areas in LNs were found as accuracy: 0.9656/0.9818, recall: 0.9185/0.9507, precision: 0.9248/0.8749, F-score: 0.8727/0.8430 by AI, at 13x ZL, respectively. In distinguishing PTC and IDC; sensitivity, specificity, recall, precision, F-score and accuracy were found to be higher than 0.97, at 13x ZL. Conclusion AI recognizes metastases with considerable accuracy. With the perfection of AI algorithms, it can be expected that AI will replace the pathologist in time-consuming tasks in the near future, providing the precious time we need for more sophisticated and enjoyable tasks.

Key words: pathology, artificial intelligence, whole slide imaging, region of interest, lymph node, metastasis