A deep-learning model, developed on cytology images of hydrotorax and ascites from 57220 cases with cancer of unknown primary (CUP) managed at four tertiary hospitals, shows high accuracy in tumour origin prediction and presents prognostic value when patient treatment is consistent with the cancer origin predicted by the model.

This method for tumour origin differentiation using cytological histology (TORCH) can serve as an effective tool in differentiation between malignancy and benign lesions, and furthermore as an auxiliary proof of concept for tumour origin prediction using cytological images. The high technical performance and potential clinical benefits of TORCH warrant further investigation in prospective randomised studies according to a group of authors from China, who published the findings on 16 April 2024 in the Nature Medicine.

Immunohistochemistry is usually applied as a key mean of predicting probable origin of CUP; however, less than 30% of CUP cases can be pinpointed by cocktails of approximately 20 different immunostaining subunits. Among patients with newly diagnosed CUP, a substantial portion present with pleural or peritoneal metastasis. Cytological examination by peritoneal or pleural fine-needle aspiration is usually used as a key method in the diagnosis of thoracoabdominal metastases. Most often, however, pathologists can visually distinguish adenocarcinoma from squamous carcinoma on cytology smears, but not the origin of the tumour cells.

Computerised analysis based on deep convolutional neural networks has recently been increasingly applied as an auxiliary technique in the field of pathological diagnosis. Artificial intelligence (AI) showed potential benefits as a diagnostic assistive tool for CUP origin prediction using whole-slide images. However, employment of AI in the prediction of cancer origin using cytological images from hydrothorax and ascites has not been investigated. In this study, the authors aimed to establish a diagnostic model to predict the broad cancer origins in patients with cancer and hydrothorax or ascites metastasis using cytological images.

Leveraging cytological images from 57,220 cases at four tertiary hospitals, the study team developed a deep-learning method that can identify malignancy and predict tumour origin in both hydrothorax and ascites. They examined its performance on three internal (n = 12,799) and two external (n = 14,538) testing sets. In both internal and external testing sets, TORCH achieved area under the receiver operating curve values ranging from 0.953 to 0.991 for cancer diagnosis and 0.953 to 0.979 for tumour origin localisation.

TORCH accurately predicted primary tumour origins, with a top-1 accuracy of 82.6% and top-3 accuracy of 98.9%. Compared with results derived from pathologists, TORCH showed better prediction efficacy (1.677 versus 1.265, p < 0.001), enhancing junior pathologists’ diagnostic scores significantly (1.326 versus 1.101, p < 0.001). Patients with CUP whose initial treatment protocol was concordant with TORCH-predicted origins had better overall survival than those who were administrated discordant treatment (27 versus 17 months, p = 0.006).

It is a challenging task to identify the origins of metastatic free tumour cells using limited clinical information and cytological images. This model achieved robust performance across five testing sets and outstanding accuracy versus a group of four pathologists. The study underscores the potential of TORCH as a valuable ancillary tool in clinical practice, although further validation in randomised studies is warranted.

This work was supported by grants from the National Natural Science Foundation of China, the National Key Research and Development Program of China, the Program for Changjiang Scholars and Innovative Research Team in University in China, the National Natural Science Cultivation Foundation of Tianjin Cancer Hospital and the Tianjin Key Medical Discipline (Specialty) Construction Project.

Reference

Tian F, Liu D, Wei N, et al. Prediction of tumor origin in cancers of unknown primary origin with cytology-based deep learning. Nature Medicine; Published online 16 April 2024. DOI: https://doi.org/10.1038/s41591-024-02915-w

Source