lacss 0.1.3

Cell segmentation and tracking

LACSS

LACSS is a model for single-cell segmentation and cell-lineage tracking

Ref: https://www.nature.com/articles/s42003-023-04608-5

As a segmentation model, it can work similar to other instance segmentation models such as MaskRCNN. However, it also support end-to-end training with very weak supervisions: e.g (a) image-level segmentation, and (b) location-of-interests (LOIs). These annotatins are chosen because they can often be produced progammably using simple unsupervised algorithms from experimental data. Our goal is to build a streamlined annotation-training pipeline that requires no manual input from humans.

The segmentation model is used for down-stream cell-tracking task. The tracking logic is based on SMC (sequential Monte Carlo).

This particular version of LACSS is build on Jax framework. Both the segmentation model and the tracking logic heavily utilize the composable transformation facility provided by JAX.