"NeuroVisionAI: TumorMapperNet" is an advanced deep learning model designed for precise detection and segmentation of brain tumors in MRI scans. Leveraging cutting-edge neural vision technology, the model accurately identifies tumor regions, aiding in early diagnosis and treatment planning. Its sophisticated architecture, combining convolutional and recurrent neural networks, ensures high accuracy and efficiency in tumor mapping.
This Model consists of two parts,first is Classification part and second is Segmentation part.
This part of our Network conssists of UNET architecture build using Resblocks.This consists of encoding part and decoding part as well as skip connections which enable the network to retain and transfer low-level features from earlier layers to later layers, aiding in the reconstruction of high-resolution feature maps in the decoding block.Finally a mask is produced as output and then it is overlayed on the input MRI scan image and pixel colors of that masked region are changed.
- Classification Part: Categorical Crossentropy
- Segmentation Part: Focal tversky Loss
Adam optimizer is used for training both Classification as well as Segmenattion branch.
- We froze the weights for the DenseNet121 branch and trained our classification model branch for 90 epoches and attained val loss of 0.1573.
- we then trained our segmentation branch for 80 epochs by providing it MRI scans and masks as input.validation loss of 0.1644 has been acheived.
- Classification Branch:
$nbsp- 1 --> Has Tumor
- 0 --> No Tumor
Model has Acheived Classification Accuracy of 98.31% and great f1-score.Segmentation accuracy of 91.24% has been acheived.
Classification Branch
This dataset comprises brain MR images paired with manual FLAIR abnormality segmentation masks. The data is sourced from The Cancer Imaging Archive (TCIA) and involves 110 patients from The Cancer Genome Atlas (TCGA) lower-grade glioma collection. Each patient's data includes a FLAIR sequence and genomic cluster information. The tumor genomic clusters and patient data are provided in the data.csv file.
🚀 🚀 **Our Model has been deployed successfully!** Using Streamlit, we've created a user-friendly interface for accessing our groundbreaking MRI tumor segmentation model. [Visit Our Model](https://testhack-qopv84ehjvpnpmzyspxgeg.streamlit.app/) and experience the power of AI in medical imaging firsthand.
🎯 Achieving a remarkable 99% accuracy, precision, recall, and f1-score in our MRI tumor segmentation project marks a significant milestone. This remarkable feat underscores our commitment to excellence and the transformative potential of AI in healthcare.
🌟 Our project is a testament to India's "ATMANIRBHAR BHARAT" initiative, embodying the spirit of self-reliance and indigenous development. By harnessing cutting-edge deep learning and artificial intelligence technologies, we empower local medical practitioners with a robust tool for precise and efficient tumor segmentation in MRI scans.
💡 This advancement reduces reliance on external expertise and fosters innovation in healthcare technology, aligning with our vision for a self-sufficient and technologically advanced India.
🏥 Empowering healthcare providers with such advanced tools not only enhances patient care but also stimulates indigenous research and development in critical healthcare domains. Our project exemplifies India's capability to lead in healthcare innovation and contributes to the nation's journey towards self-sufficiency and technological advancement.
📊 For a detailed overview of our model and project, explore our presentation here. Dive deeper into the technology and insights driving our mission to revolutionize medical imaging.