Abstract 333P
Background
Translational Renal Cell Carcinoma (TRCC) is a rare subtype of renal cell carcinoma that often mimics the pathological morphology of clear cell renal cell carcinoma (ccRCC) and papillary renal cell carcinoma. Historically, distinguishing TRCC from these subtypes using hematoxylin and eosin (HE) stained morphology has been challenging. The advent of large-scale deep learning models now offers a promising solution to this issue.
Methods
This study involved a retrospective screening of approximately 8,000 renal cell carcinoma cases from three medical centers, including 80 TRCC and 80 ccRCC cases represented in HE stained slide scans. After cropping images, segmenting into patches, removing blank images, and standardizing color, the data was divided into a training set (n=100) and a validation set (n=60). A Transformer model was employed for training and feature extraction. Subsequently, a machine learning model capable of distinguishing TRCC from ccRCC was developed using the features derived from the Transformer model, employing algorithms such as ExtraTrees, Random Forests, KNN, XGBoost, and LightGBM.
Results
The processing of approximately 1.5 million patch patterns from the pathological images enabled significant differentiation between TRCC and ccRCC based on Transformer-derived features. The area under the curve (AUC) for the training set reached 0.932, while the validation set achieved an AUC of 0.869. Among the machine learning models tested, the ExtraTrees model, when integrated with Transformer model features, demonstrated superior performance in accurately identifying TRCC.
Conclusions
This multicenter study successfully developed and validated a deep learning-based model for the identification of TRCC using pathological omics data. The findings highlight the potential of integrating advanced machine learning techniques with traditional pathological analysis to enhance diagnostic accuracy for rare cancer subtypes.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.