This year we worked with colleagues from the National Cancer Insitutes (NCI) and their collaborators at Intellectual Ventures-Global Good, the Bill and Melinda Gates Foundation, and Mobile ODT. We validated a Convolutional Neural Network (CNN) based DL algorithm trained and tested on a sample of cervical images selected from a large archive of digitized cervical images from a 9,450-woman, population-based longitudinal cohort (ages 18-92) acquired by the NCI in Guanacaste, Costa Rica. The cohort provided cervical training/validation images and clinical endpoints. AVE generates a severity score (0 to 1); the positivity threshold can be adjusted to balance precancer detection with numbers of women unnecessarily treated. AVE screening very accurately identified prevalent precancer/cancer (AUC = 0.95). Applied to enrollment cervical images, it outperformed standard screening tests (clinician interpretation of the same cervical images, Pap smears, and even HPV testing) in predicting cumulative risk of precancer/cancer. AVE provides sensitive screening with minimal clinical training or cost. Overtreatment still must be addressed by further improvements, unless lower sensitivity is accepted. Other the cervicographic images, we also devoted efforts toward analysis DL-based classification of histology images acquired from tissue biopsy slides. This year a deep learning (DL)-based nuclei segmentation approach was investigated based on gathering localized information through the generation of superpixels using a simple linear iterative clustering algorithm and training with a CNN-DL algorithm. The proposed approach was evaluated on a dataset of 133 digitized histology images and achieved an overall nuclei detection (object-based) accuracy of 95.97%, with demonstrated improvement over imaging-based and clustering-based benchmark techniques. Building on our previous research, we also introduced an automated localized, fusion-based algorithm to classify squamous epithelium in histology images into Normal, and various Cervical Intraepithelial Neoplasia (CIN) grades, viz. CIN1, CIN2, and CIN3. The approach partitioned the epithelium into 10 segments. Image analysis algorithms were used to extract features from each segment. The features were then used to classify each segment and these results were subsequently fused to classify the whole epithelium. This research applies CNN-DL algorithm and resulted in an increased classification accuracy to 77.25%.
|Sornapudi, Sudhir; Stanley, Ronald Joe; Stoecker, William V et al. (2018) Deep Learning Nuclei Detection in Digitized Histology Images by Superpixels. J Pathol Inform 9:5|