AUTHOR=Siarov Jan , Siarov Angelica , Kumar Darshan , Paoli John , Mölne Johan , Neittaanmäki Noora 

TITLE=Deep learning model shows pathologist-level detection of sentinel node metastasis of melanoma and intra-nodal nevi on whole slide images

JOURNAL=Frontiers in Medicine

VOLUME=Volume 11 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2024.1418013

DOI=10.3389/fmed.2024.1418013

ISSN=2296-858X

ABSTRACT=The presence of nodal metastasis (NM) in sentinel node biopsies (SNB) is a crucial part in melanoma staging. Further, an intra-nodal nevus (INN) may be misclassified as NM. There is high discordance in assessing SNB positivity which may lead to false staging. The use of digital whole slide imaging enables the implementation of artificial intelligence (AI) in digital pathology.We assessed the capability of AI in detection of NM and INN in SNBs.: In total, 485 hematoxylin and eosin whole slide images (WSIs) including NM and INN from 196 SNBs were collected and divided into training (279 WSIs), validation (89 WSIs) and test sets (117 WSIs). The deep learning model was trained with 5,956 manual pixelwise annotations. The test set was assessed by the AI and three blinded dermatopathologists. Immunohistochemistry served as the reference standard. Results: The AI model showed excellent performance with an area under the curve receiver operating characteristic (AUC) of 0.965 for detection of NM. For comparison, AUC for NM detection among the dermatopathologists varied between 0.94 and 0.98. For detection of INN, AUC was lower for both AI (0.781) and the dermatopathologists (range 0.63-0.79). Discussion: To conclude, the deep learning AI model showed excellent accuracy in detection of NM and dermatopathologist-level performance in detection of both NM and INN. Importantly, the AI model showed potential in differentiating between these two entities. Further validation is still warranted.