Implementation

of

LMBiS-Net

Implementation of Abbasi et al’s Paper: “LMBiS-Net: A Lightweight Multipath ​Bidirectional Skip Connection based CNN for Retinal Blood Vessel Segmentation”

Milad Farazian, Charlie Floeder, Rizq Khateeb, Harshit Shah, Yash Sharma

Original Paper

PROJECT GOAL

To implement the LMBiS-Net model and confirm the ​findings presented in “LMBiS-Net: A Lightweight ​Multipath Bidirectional Skip Connection based CNN for ​Retinal Blood Vessel Segmentation.” A second goal is ​to apply our implementation to a different dataset that ​is not used in the paper.

PROJECT IMPORTANCE

LMBiS-Net’s primary benefit is it provides an accurate ​retinal blood vessel segmentation model that is more ​computationally efficient compared to current state-of-​the-art models. Retinal blood vessel segmentation is ​beneficial for early detection and treatment of retinal ​diseases. This model can significantly reduce the ​amount of time ophthalmologists spend manually ​identifying retinal vessels. Additionally, the model can ​reduce human error in that task. Retinal diseases are a ​major cause of visual impairment and blindness. ​Studies show that 5%-20% of the global population ​ages 40+ have retinal disorders. Examining retinal ​vessels can provide important information regarding ​the underlying medical conditions leading to retinal ​diseases.


LM​BiS-NET

LMBiS-Net is a CNN consisting of three encoder ​blocks, a bottleneck layer, and three decoder blocks. ​The network uses multipath feature extraction blocks ​and incorporates bidirectional skip connections for the ​information flow between the encoders and decoders.


MULTI-PATH FEATURE EXTRACTION BLOCK

This is a key component to LMBiS-Net, which introduces feature diversity into the model and ​helps prevent overfitting. By processing the data using different sized convolutions, the network ​can learn to capture both low-level and high-level features which is important for blood vessel ​segmentation.


PROJECT CONTRIBUTION

We created the first publicly available implementation of ​LMBiS-Net along with code to augment retinal images to ​increase the size of training datasets. Our findings add ​credibility to the original paper’s claims that LMBiS-Net is ​a computationally efficient and accurate state-of-the-art ​retinal blood vessel segmentation model.


RESULTS

Google Colab