A Transfer Learning Radiomics Nomogram for Preoperative Prediction of Borrmann Type IV Gastric Cancer From Primary Gastric Lymphoma

Abstract

ObjectiveThis study aims to differentiate preoperative Borrmann type IV gastric cancer (GC) from primary gastric lymphoma (PGL) by transfer learning radiomics nomogram (TLRN) with whole slide images of GC as source domain data.Materials and MethodsThis study retrospectively enrolled 438 patients with histopathologic diagnoses of Borrmann type IV GC and PGL. They received CT examinations from three hospitals. Quantitative transfer learning features were extracted by the proposed transfer learning radiopathomic network and used to construct transfer learning radiomics signatures (TLRS). A TLRN, which integrates TLRS, clinical factors, and CT subjective findings, was developed by multivariate logistic regression. The diagnostic TLRN performance was assessed by clinical usefulness in the independent validation set.ResultsThe TLRN was built by TLRS and a high enhanced serosa sign, which showed good agreement by the calibration curve. The TLRN performance was superior to the clinical model and TLRS. Its areas under the curve (AUC) were 0.958 (95% confidence interval [CI], 0.883–0.991), 0.867 (95% CI, 0.794–0.922), and 0.921 (95% CI, 0.860–0.960) in the internal and two external validation cohorts, respectively. Decision curve analysis (DCA) showed that the TLRN was better than any other model. TLRN has potential generalization ability, as shown in the stratification analysis.ConclusionsThe proposed TLRN based on gastric WSIs may help preoperatively differentiate PGL from Borrmann type IV GC.Borrmann type IV gastric cancer, primary gastric lymphoma, transfer learning, whole slide image, deep learning.

Classification of tumor from computed tomography images: A brain-inspired multisource transfer learning under probability distribution adaptation

“…To verify the superiority of the proposed AMSF-L1ELM, several state-of-the-art methods are used for comparison in two validation cohorts. Representative methods include the classic radiomics signature (RS) (Feng et al, 2021) based on radiomics features, a clinical model (CM) based on subjective signs and clinical information (Feng et al, 2021) From these results in Table 2, we can obtain the following insightful observations.…”

“…Considering clinical practice, doctors often need to use a variety of means for disease diagnosis, such as lesion screening based on multisequence CT images (such as arterial phase and venous phase) and then pathological diagnosis based on WSI images. Additionally, the effectiveness of the transfer learning model, which was based on single-source domain WSI of gastric tissue and WSI of lung tissue, was preliminarily validated according to our previous work ( Feng et al, 2021 ). Considering the above observations, this manuscript plans to select the gastric WSIs (source domain 1), lung WSIs (source domain 2) and arterial phase CT images of gastric cancer (source domain 3) as the source domain data, and proposes a multisource transfer learning feature extraction network, as shown in Figure 3 .…”

Development and validation of a predictive model combining clinical, radiomics, and deep transfer learning features for lymph node metastasis in early gastric cancer

“…However, acquiring a large number of medical images is difficult ( 20 ). Due to a pre-trained CNN known as “transfer learning (TL)” can be used to minimize overfitting with a small training size, TL has gradually been used in various medical image analysis domains in recent years ( 21 , 22 ). TL increases model performance in target tasks by transferring previously learned features from source tasks.…”

“…TL increases model performance in target tasks by transferring previously learned features from source tasks. A previous study found that a TL radiomics nomogram based on gastric whole slide images can assist in distinguishing primary gastric lymphoma from Borrmann type IV GC ( 22 ). In addition, Linlin et al ( 21 ) developed a convenient model based on deep learning-based radiomics characteristics to differentiate brain abscess from cystic glioma.…”

A brain-like classification method for computed tomography images based on adaptive feature matching dual-source domain heterogeneous transfer learning

“…However, the size of datasets in the medical field is often small, and CNN is prone to over-fitting under the condition of small samples. To improve the effect of CNN under small samples, scholars have introduced transfer learning into CNN ( Tan et al, 2018 ; Feng et al, 2021 ). By imitating the mechanism of drawing inferences about other cases from one instance of the human brain, transfer learning uses the knowledge learned from a large source domain data to help the learning of the target task.…”