Classifying Malignancy in Prostate Glandular Structures from Biopsy Scans with Deep Learning.

Histopathological classification in prostate cancer remains a challenge with high dependence on the expert practitioner. We develop a deep learning (DL) model to identify the most prominent Gleason pattern in a highly curated data cohort and validate it on an independent dataset. The histology images are partitioned in tiles (14,509) and are curated by an expert to identify individual glandular structures with assigned primary Gleason pattern grades. We use transfer learning and fine-tuning approaches to compare several deep neural network architectures that are trained on a corpus of camera images (ImageNet) and tuned with histology examples to be context appropriate for histopathological discrimination with small samples. In our study, the best DL network is able to discriminate cancer grade (GS3/4) from benign with an accuracy of 91%, F1-score of 0.91 and AUC 0.96 in a baseline test (52 patients), while the cancer grade discrimination of the GS3 from GS4 had an accuracy of 68% and AUC of 0.71 (40 patients).

EPB41L5 is Associated With the Metastatic Potential of Low-grade Pancreatic Neuroendocrine Tumors.

BACKGROUND/AIM: Low-grade pancreatic neuroendocrine tumors (LG-PNETs) behave unpredictably. The aim of the study was to identify biomarkers that predict PNET metastasis to improve treatment selection. PATIENTS AND METHODS: Five patients with primary non-metastatic LG-PNETs, six with primary LG-PNETs with synchronous or metachronous metastases (M-PNETs), and six metastatic to liver LG-PNETs (ML-PNETs) from the group of six M-PNET patients were selected. RNA data were normalized using iterative rank-order normalization. Student's t-test identified differentially-expressed genes in LG-PNETs versus M-PNETs. A 2-fold difference in expression was considered to be significant. Results were validated with an independent dataset of LG-PNETs and metastatic LG-PNETs. RESULTS: Overall, 195 genes had a >2-fold change (in either direction). A total of 29 genes were differentially overexpressed in M-PNETs. Erythrocyte membrane protein band 4.1-like 5 (EPB41L5) had a 2.07-fold change increase in M-PNETs and the smallest p-value. EPB41L5 was not statistically different between M-PNETs and ML-PNETs. EPB41L5 differential expression between primary and metastatic LG-PNETs was confirmed by immunohistochemistry. CONCLUSION: These results support further investigation into whether EPB41L5 is a biomarker of PNETs with high risk for metastases.

IMP3 Immunoreactivity is More Sensitive Than AMACR in Detecting Dysplastic Epithelium and Early Adenocarcinoma in Barrett Esophagus.

CONTEXT: α-methylacyl coenzyme A racemase (AMACR) and insulin-like growth factor-II mRNA-binding protein 3 (IMP3) are 2 markers helpful in detecting difficult cases of dysplasia in Barrett esophagus (BE). However, no comparison studies have been performed to assess their performance in the same patient population. OBJECTIVES: The aim of our study was to compare the immunohistochemical expression of IMP3 and AMACR in dysplastic lesions and early adenocarcinoma (EAC) arising in BE and evaluate their sensitivity and specificity. DESIGN: A total of 98 cases [BE negative for dysplasia, n=24; indefinite for dysplasia (BE-IND), n=18; low-grade dysplasia (LGD), n=24; high-grade dysplasia (HGD), n=16; and EAC, n=16] were immunostained for AMACR and IMP3 and evaluated for the degree, the extent, and the intensity of staining. RESULTS: No immunoreactivity for AMACR or IMP3 was observed in all 24 cases of BE negative for dyplasia. One of 18 (5.5%) cases of BE-IND was positive for IMP3, but all were negative for AMACR. AMACR and IMP3 were positive in 16.7% versus 41.7 % of the cases with BE-LGD, 25% versus 62.5% of BE-HGD, and 62.5% versus 93.7% of EAC, respectively. The sensitivity of AMACR and IMP3 for the detection of dysplasia in BE is 16.7% and 41.7% for LGD, 25% and 62.5% for HGD, and 62.5% and 93.7% in EAC, respectively. The specificity is 100% for both markers. In addition, a comparison of the intensity of reactivity shows a better result with IMP3 (36/98, 36.7%) than with AMACR (18/98, 18.4%) (P<0.001). CONCLUSIONS: IMP3 has a similar specificity, but a better sensitivity, intensity, and extent of reactivity in comparison with AMACR, and may be used as an alternative to AMACR, in support of the diagnosis of BE-dysplasia and EAC.