Blasto-Net: An Explainable Multi-Task Learning for Blastocyst Segmentation, Grading, and Implantation Prediction

Researchers have proposed three new deep learning models for medical applications: Blasto-Net for blastocyst analysis, FAROS for surgical scene understanding, and a multi-task framework for intracranial aneurysm classification.

Blasto-Net, a multi-task deep learning model, performs simultaneous segmentation, grading, and implantation prediction for blastocysts in IVF treatment^[1]. It achieves Dice scores of 94.93%, 91.60%, and 88.82% for ICM, ZP, and TE, respectively, and an implantation F1-score of 80.0% on the HMC blastocyst dataset. The model employs an EfficientNet-B3 encoder with a UNet-style decoder enhanced by CBAM and EAAM modules^[1]. Meanwhile, researchers have introduced FAROS, a framework for surgical scene understanding that addresses annotation granularity mismatch. FAROS combines zero-shot segmentation-based mask propagation with optical flow estimation to generate temporally consistent dense pseudo labels^[2]. It has been validated on the DAVIS 2017 benchmark and tested on GraSP, MISAW, and AutoLaparo benchmarks, showing robust propagation beyond the surgical domain. Additionally, a new multi-task framework has been developed for classifying and segmenting intracranial aneurysms using tri-axial ROI and multi-task learning^[3]. This framework simultaneously performs multi-label classification, multi-class aneurysm segmentation, and multi-class vessel segmentation, addressing the limitations of existing binary detection methods.