Seeking Researcher for Multimodal Deep Learning Model (Lumbar Spine MRI + Clinical Data for Osteoporosis Prediction)
Budget: $750 – $1,500 USD
I am looking for an experienced researcher or data scientist with expertise in deep learning for medical imaging to help analyze my dataset and write the Methods and Results sections for a scientific manuscript.
Project background:
Goal: Develop a multimodal fusion deep learning model combining MRI (T1- and T2-weighted lumbar spine images) with structured clinical data to predict lumbar spine bone mineral density (BMD) or diagnose osteoporosis.
Definition of Osteoporosis: OP is defined as a T-score ≤ −2.5 standard deviations from the young adult mean, measured by DXA.
Dataset:
Patient cohort: Approximately 300 patients, resulting in around 1,000 vertebral segments after excluding segments with compression fractures (each vertebra considered an independent sample).
Imaging: For each lumbar vertebra (L1, L2, L3, L4), we have one T1-weighted sagittal image and one T2-weighted sagittal image.
Labels: Each vertebra has corresponding BMD value and binary osteoporosis classification (yes/no).
Clinical variables:
Demographics: age, sex, body weight, body height
Past medical history: diabetes mellitus, hypertension, dyslipidemia, coronary artery disease (CAD)
Lifestyle factors: smoking history, alcohol intake history
Target model architecture: Similar to published multimodal approaches combining CNN-based imaging features (e.g., EfficientNetB7 or equivalent) with feedforward neural networks for clinical data, using feature concatenation before classification
.
Deliverables:
Data preprocessing, including normalization, augmentation, and patient-level data split.
Model implementation with multimodal fusion (MRI + clinical data).
Training, validation (K-fold or hold-out), and performance evaluation (accuracy, AUC, sensitivity, specificity, F1-score).
Optimization of hyperparameters (e.g., via Optuna or genetic algorithm).
Drafting of the Methods and Results sections of the manuscript, following academic style and including tables/figures.
Requirements:
Proven experience in medical imaging deep learning (preferably MRI) and multimodal model development.
Familiarity with Python, PyTorch or TensorFlow/Keras, and medical image processing (DICOM/NIfTI).
Strong academic writing skills in English.
Ability to produce reproducible code and clear documentation.
If you have relevant publications or prior similar projects, please include them in your proposal.
Project background:
Goal: Develop a multimodal fusion deep learning model combining MRI (T1- and T2-weighted lumbar spine images) with structured clinical data to predict lumbar spine bone mineral density (BMD) or diagnose osteoporosis.
Definition of Osteoporosis: OP is defined as a T-score ≤ −2.5 standard deviations from the young adult mean, measured by DXA.
Dataset:
Patient cohort: Approximately 300 patients, resulting in around 1,000 vertebral segments after excluding segments with compression fractures (each vertebra considered an independent sample).
Imaging: For each lumbar vertebra (L1, L2, L3, L4), we have one T1-weighted sagittal image and one T2-weighted sagittal image.
Labels: Each vertebra has corresponding BMD value and binary osteoporosis classification (yes/no).
Clinical variables:
Demographics: age, sex, body weight, body height
Past medical history: diabetes mellitus, hypertension, dyslipidemia, coronary artery disease (CAD)
Lifestyle factors: smoking history, alcohol intake history
Target model architecture: Similar to published multimodal approaches combining CNN-based imaging features (e.g., EfficientNetB7 or equivalent) with feedforward neural networks for clinical data, using feature concatenation before classification
.
Deliverables:
Data preprocessing, including normalization, augmentation, and patient-level data split.
Model implementation with multimodal fusion (MRI + clinical data).
Training, validation (K-fold or hold-out), and performance evaluation (accuracy, AUC, sensitivity, specificity, F1-score).
Optimization of hyperparameters (e.g., via Optuna or genetic algorithm).
Drafting of the Methods and Results sections of the manuscript, following academic style and including tables/figures.
Requirements:
Proven experience in medical imaging deep learning (preferably MRI) and multimodal model development.
Familiarity with Python, PyTorch or TensorFlow/Keras, and medical image processing (DICOM/NIfTI).
Strong academic writing skills in English.
Ability to produce reproducible code and clear documentation.
If you have relevant publications or prior similar projects, please include them in your proposal.