.png){kind=icon}
.png){kind=icon}
Return
Ultra-sensitive ovarian cancer diagnosis using deep residual networks and multimodal SPR signal signatures
.png){kind=icon}
PRE
AI
R
M
S
B
DOI：10.1007/s11082-026-09047-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Early diagnosis of ovarian cancer is also a significant clinical problem because the concentration of biomarkers at an early stage is extremely low. The study suggests a unified framework that includes the invention of a Surface Plasmon Resonance (SPR)-based biosensing platform and a deep residual neural network (ResNet) in the early detection of ovarian cancer. The constructed SPR sensor, which is functionalized with monoclonal antibodies against CA125, exhibited a distinct and almost linear response in the low concentration area, generating a stable response within the range of 0 to 20 U/mL with responses rising between 0 and 330 RU, a strong indicator of high sensitivity and reliability. Following preprocessing, the signal-to-noise ratio was 10 (raw) and 23 (normalized), which is much better and allows for more quality and useful data. The convergence of the proposed ResNet model was steady with a classification accuracy of 98% after 50 training epochs. Separability of features also increased smoothly with depth of network, starting with 0.20 at the input layer and 1.45 at the last layer, which proves successful hierarchical feature learning. The proposed method also reached higher performance in comparison studies than the conventional methods, with the 97% accuracy relative to 78% (threshold), 85% (PCA), and 92% (CNN). When tested on an independent held-out set, it achieved a 97% accuracy score, 98% sensitivity, and 96% specificity. Under the analysis of the importance of features, CA125 binding amplitude and kinetic slopes were found to be the biggest contributors.
Keywords:
Surface plasmon resonance
Ovarian cancer detection
Biosensor
Deep residual network
Biomarker analysis
Machine learning
Journal
IF:
4
Papers:
9.8K
Citations:
1.8W
Organization
Cited Papers