A Calibrated Multiexit Neural Network for Detecting Urothelial Cancer Cells.

Deep convolutional networks have become a powerful tool for medical imaging diagnostic. In pathology, most efforts have been focused in the subfield of histology, while cytopathology (which studies diagnostic tools at the cellular level) remains underexplored. In this paper, we propose a novel deep learning model for cancer detection from urinary cytopathology screening images. We leverage recent ideas from the field of multioutput neural networks to provide a model that can efficiently train even on small-scale datasets, such as those typically found in real-world scenarios. Additionally, we argue that calibration (i.e., providing confidence levels that are aligned with the ground truth probability of an event) has been a major shortcoming of prior works, and we experiment a number of techniques to provide a well-calibrated model. We evaluate the proposed algorithm on a novel dataset, and we show that the combination of focal loss, multiple outputs, and temperature scaling provides a model that is significantly more accurate and calibrated than a baseline deep convolutional network.

CapsoCam SV-1 Versus PillCam SB 3 in the Detection of Obscure Gastrointestinal Bleeding: Results of a Prospective Randomized Comparative Multicenter Study.

BACKGROUND AND STUDY AIM: Newer capsule with a panoramic viewing mode is available and might increase the detection rate of bleeding lesions in patients with obscure gastrointestinal bleeding (OGIB). Furthermore, an improved patient acceptance rate is expected. MATERIALS AND METHODS: In a randomized prospective comparative multicenter study, patients with OGIB were included and examined either with CapsoCam SV-1 or with PillCam SB 3. Detection of bleeding lesions, transit, and evaluation time and adverse events were evaluated. Physicians were interviewed about their experience with both capsules and the evaluation software. A detailed subject questionnaire analyzed acceptance of each capsule. Follow-up was 3 months. RESULTS: In total, 181 patients with OGIB were recruited into the study. After exclusion of 28 patients 153 patients were randomized and CapsoCam SV-1 (n=78) or PillCam SB 3 (n=75) was administered. CapsoCam SV-1 detected more cases of bleeding (31/79, diagnostic yield 39.7%) compared with PillCam SB 3 (26/75, diagnostic yield 34.6%, NS). Transit time of both capsules was not different. Evaluation time with PillCam SB 3 was superior to CapsoCam SV-1 (27 vs. 40 min, P=0.01). In total, 95% of the physicians were satisfied with each capsule system and evaluation software. The acceptance rate of the patients to retrieve the CapsoCam SV-1 was high. Adverse events/serious adverse events were 17.9%/1.3% with CapsoCam SV-1 and 16%/0% with PillCam SB 3. Rebleeding rate was 28.75% within 3 months. CONCLUSIONS: CapsoCam SV-1 detected more lesions; however, relevant bleeding sources were visualized by both capsules. Physician's satisfaction was high with both capsule systems and evaluation software. Patient's acceptance with CapsoCam SV-1 was unexpectedly high. Serious adverse events were 0% with PillCam SB 3 and 1.3% with CapsoCam SV-1.

Expression of luteal estrogen receptor, interleukin-1, and apoptosis-associated genes after PGF2alpha administration in rabbits at different stages of pseudopregnancy.

The dynamic expression for estrogen receptor subtype-1 (ESR1), interleukin-1beta (IL1B), and apoptosis-associated genes, as well as nitric oxide synthase activity, were examined in corpora lutea (CL) of rabbits after prostaglandin F(2alpha) (PGF(2alpha)) administration on either day 4 or day 9 of pseudopregnancy. By reverse transcriptase polymerase chain reaction, the steady-state level of ESR1 transcript was lower (P < 0.01) and that of anti-apoptotic B-cell CLL/lymphoma 2 (BCL2) -like 1 (BCL2L1) was greater in day 4 (P < 0.01) than in day 9 CL. Western blot analysis revealed that BCL2-associated X protein (BAX) abundance was greater in day 4 (P < 0.01) than in day 9 CL, whereas BCL2L1 protein was undetectable at both luteal stages. After PGF(2alpha), ESR1 transcript decreased (P < 0.01) in day 9 CL, whereas IL1B mRNA showed a transitory increase (P < 0.01) at both stages. The pro-apoptotic tumor protein p53 (TP53) gene had diminished (P < 0.01) on day 4 and on day 9 after a transitory increase (P < 0.01), whereas the BAX/BCL2L1 expression ratio increased (P < 0.01) in day 9 CL 24 h after treatment. Following PGF(2alpha), TP53 protein increased (P < 0.01) at both luteal stages, and BAX decreased (P < 0.01) in day 4 CL but increased (P < 0.01) 24 h later in day 9 CL; BCL2L1 became detectable 6 h later in day 4 CL. Nitric oxide synthase activity temporarily increased (P < 0.01) following PGF(2alpha). These findings suggest that PGF(2alpha) regulates luteolysis by ESR1 mRNA down-regulation and modulation of pro- and anti-apoptotic pathways in CL that have acquired a luteolytic capacity.