AUTHOR=Li Chunyu , Deng Ming , Zhong Xiaoli , Ren Jinxia , Chen Xiaohui , Chen Jun , Xiao Feng , Xu Haibo 

TITLE=Multi-view radiomics and deep learning modeling for prostate cancer detection based on multi-parametric MRI

JOURNAL=Frontiers in Oncology

VOLUME=Volume 13 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1198899

DOI=10.3389/fonc.2023.1198899

ISSN=2234-943X

ABSTRACT=This study aims to develop an imaging model based on multi-parametric MR image for distinguishing between prostate cancer (PCa) and prostate hyperplasia. 236 subjects were enrolled and divided into training and test sets for models construction. Firstly, a multi-view radiomics modelling strategy was designed, in which different combinations of radiomics features categories (original, LoG and wavelet) were compared to obtain the optimal input feature sets. Minimum-redundancy maximum-relevance (mRMR) selection and least absolute shrinkage selection operator (LASSO) were used for features reduction and next logistic regression method was used for models construction. Then, a Swin Transformer architecture was designed and trained using transfer learning techniques to construct the deep learning models (DL). Finally, the constructed multi-view radiomics and DL models were combined and compared for model selection and nomogram construction. The prediction accuracy, consistency and clinical benefit were comprehensively evaluated in model comparison. Optimal input feature set were found when LoG and wavelet features were combined, while 22 and 17 radiomics features in this set were selected to construct the ADC and T2 multi-view radiomics model, respectively. ADC and T2 DL models were built by transfer learning from a large amount of natural images to our relative small sample of prostate images. All individual and combined models showed good predictive accuracy, consistency and clinical benefit. Compared with using only ADC-based model, adding T2-based model into the combined model would reduce model predictive performance. ADCCombinedScore model showed best predictive performance among all, and was transformed into a nomogram for better use in clinic. The constructed models in our study can be used as a predictor in the differentiating of PCa and BPH, thus helping clinicians to make better clinical treatment decisions and reducing unnecessary prostate biopsies.