Learning Under Extreme Data Scarcity: Subject-Level Evaluation of Lightweight CNNs for fMRI-Based Prodromal Parkinsons Detection

2026-03-03 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, AI in Healthcare · Depth: Intermediate, quick

Summary

A study on prodromal Parkinson's disease detection using resting-state fMRI data from 40 subjects (20 prodromal cases, 20 controls) investigates deep learning under extreme data scarcity. It fine-tunes ImageNet-pretrained convolutional neural networks, including VGG19, Inception V3, Inception ResNet V2, and MobileNet V1. The research highlights that common image-level data splits lead to information leakage and inflated accuracy, whereas a strict subject-level split reveals test accuracies between 60% and 81%. MobileNet V1, a lightweight architecture, consistently outperforms deeper models like VGG19 under subject-level evaluation, suggesting that evaluation strategy and model capacity are more critical than architectural depth in low-data scenarios.

Key takeaway

For AI Scientists developing deep learning models with scarce neuroimaging data, you must enforce strict subject-level data splits to prevent information leakage and obtain realistic performance metrics. Your focus should be on model capacity and robust evaluation strategies rather than simply increasing architectural depth, as lightweight models like MobileNet V1 can offer superior generalization in these challenging low-data regimes.

Key insights

Subject-level data splitting and model capacity are crucial for reliable deep learning evaluation in extreme data scarcity.

Principles

• Avoid image-level data splits in neuroimaging.
• Model capacity impacts generalization in low-data.
• Evaluation strategy is critical for performance.

Method

Fine-tuning ImageNet-pretrained CNNs on fMRI data, comparing performance under image-level vs. strict subject-level data partitioning, and evaluating models with varying capacities.

In practice

• Use subject-level data splits for fMRI analysis.
• Consider lightweight CNNs like MobileNet V1.
• Prioritize evaluation strategy over model depth.

Topics

• Prodromal Parkinson's Detection
• fMRI Analysis
• Data Scarcity Learning
• Lightweight CNNs
• Subject-Level Evaluation

Best for: AI Scientist, Research Scientist, AI Researcher, Machine Learning Engineer, Data Scientist

Related on AIssential

• Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios — Synthetic FCD-MRIs generated by conditional networks can augment real data, reducing labeled data needs for automated detection.
• Attention-Gated Convolutional Networks for Scanner-Agnostic Quality Assessment — A hybrid CNN-Attention model effectively assesses MRI quality, generalizing across diverse scanners without retraining.
• Learning neuroimaging models from health system-scale data — Health system-scale data enables AI foundation models to achieve superior, fair, and explainable neuroimaging diagnostics.
• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• Decision processes in 3D structural MRI schizophrenia classification evaluated with saliency maps — Explainable AI methods are crucial for validating deep learning models in medical diagnostics, revealing decision plausibility beyond performance metrics.
• Multi-Class Brain Tumor Classification Using Advanced Deep Learning Models: A Comparative Study — EfficientNetB0 significantly improves multi-class brain tumor classification from MRI images, especially for subtle meningiomas, outperforming other CNNs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.