Back

Predicting the evolution of lung squamous cell carcinoma in situ using deep learning

Vigdorovits, A.; Olteanu, G.-E.; Tica, O.; Boros, M.; Pop, O.

2022-12-10 pathology
10.1101/2022.12.06.519339 bioRxiv
Show abstract

Lung squamous cell carcinoma in situ (SCIS) is the pre-invasive precursor lesion of lung squamous cell carcinoma (SCC). Only half of these lesions progress to invasive cancer, while a third undergo spontaneous regression. The ability to predict the evolution of SCIS lesions can significantly impact the management of lung cancer patients. Here, we present the use of the deep learning (DL) approach in order to predict the progression of SCIS. The dataset consisted of 112 H&E stained whole slide images (WSI) that were obtained from the Image Data Resource public repository. The data set corresponded to tumors of patients who underwent biopsies of SCIS lesions and were subsequently followed up by bronchoscopy and CT to monitor for progression to SCC. We show that a deep convolutional neural network (DCNN) can predict if a SCIS lesion will progress to SCC. The model achieved a per-tile AUC of 0.78 (SD = 0.01) on the test set, an F1 score of 0.84 (SD = 0.05), and a sensitivity of 0.94 (SD = 0.01). Class activation maps were created in order to explore how the DCNN made decisions. To our knowledge, this study is the first to demonstrate that DL has the ability to predict the evolution of SCIS from H&E WSI. DL has the potential to be used as a low-cost method that could provide prognostic information for patients with preinvasive lesions.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.