Multiple myeloma (MM) is a blood cancer caused by the abnormal expansion of plasma cells in the bone marrow. Diagnosis of multiple myeloma can be done through various methods, including CBC and counting myeloma plasma cells in aspirate slide images. Deep learning models and computer vision techniques are being used to detect MM.
- Physical simulations can be applied to raw BMP images to improve the accuracy of myeloma cell segmentation
- The goal of our work is to make the process of myeloma cell segmentation more robust
- SegPC-2021 dataset is being used for this project
- Collected from subjects suffering from Multiple Myeloma (MM) who were being diagnosed or treated at AIIMS, New Delhi, India.
- Microscopic images were captured from bone marrow aspirate slides using two cameras:
- Olympus camera with image size 2040x1536 pixels
- Nikon camera with image size 1920x2560 pixels
- The dataset contains 775 images that were stain color normalized and divided into three sets:
- train (298 images)
- validation (200 images)
- test (277 images)
- The ground truth (GT) values are available for the training and validation sets, but not for the test set
- Rescale the images to the dimensions of 1080x1440 pixels since the desired outputs are expected in this size
- Convert the given ground truth image masks into COCO annotation format
- Apply image transformations to simulate physical changes in the images
- Illumination: Create a custom trainable layer that has optimizable weights to simulate the effect of 12 light waves hitting the sample at various angles
- Contrast normalization: Enhance the visibility of myeloma cells that have varying contrast levels compared to other cells and tissues
- Morphological operations: Smooth the edges of the cells using Erosion and Dilation operations
- Gradient filters: Apply Sobel Gradient filters to enhance the edges of the cells
- Manipulating different color channels (RGB): Enhance or suppress the effect of either the red, green, or blue channel
- Blur filters: Used Gaussian & Median Blur filters to smoothen the image and reduce noise
| IMAGE TRANSFORM | NUCLEUS SEG VAL AP | CYTOPLASM SEG VAL AP | TEST MEAN IOU | MODEL WEIGHTS |
|---|---|---|---|---|
| None | 80.3993 | 44.3472 | 0.8557 | model |
| Illumination Simulation | 77.6479 | 41.4616 | 0.8469 | model |
| Contrast Normalization | 78.2172 | 44.0549 | 0.8494 | model |
| Erosion Operation | 78.7668 | 43.8360 | 0.8465 | model |
| Dilation Operation | 82.0081 | 43.7919 | 0.8586 | model |
| Sobel Gradient Filter | 74.1309 | 38.5333 | 0.8322 | model |
| Red Channel Enhanced | 79.8766 | 42.9748 | 0.8588 | model |
| Green Channel Enhanced | 79.9117 | 42.3286 | 0.8665 | model |
| Blue Channel Enhanced | 79.0457 | 43.5166 | 0.8553 | model |
| Gaussian Blur Filter | 81.7441 | 43.9569 | 0.8477 | model |
| Median Blur Filter | 81.5294 | 46.0385 | 0.8696 | model |
-
Anubha Gupta, Ritu Gupta, Shiv Gehlot, Shubham Goswami, April 29, 2021, "SegPC-2021: Segmentation of Multiple Myeloma Plasma Cells in Microscopic Images", IEEE Dataport, doi: https://dx.doi.org/10.21227/7np1-2q42 (https://dx.doi.org/10.21227/7np1-2q42).
-
The Definitive Guide to Cell Segmentation Analysis, https://blog.biodock.ai/definitive-guide-to-cell-segmentation-analysis/ (https://blog.biodock.ai/definitive-guide-to-cell-segmentation-analysis/)
-
Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick, March 20, 2017, "Mask R-CNN", arXiv, doi: https://doi.org/10.48550/arxiv.1703.06870 (https://doi.org/10.48550/arxiv.1703.06870)
-
Segmentation of Multiple Myeloma Plasma Cells in Microscopic Images (SegPC-2021), accessed March 10, 2023, https://segpc-2021.grand-challenge.org/ (https://segpc-2021.grand-challenge.org/)
-
Digging into Detectron 2 — part 1, accessed April 15, 2023, https://medium.com/@hirotoschwert/digging-into-detectron-2-47b2e794fabd (https://medium.com/@hirotoschwert/digging-into-detectron-2-47b2e794fabd)