[Paper Review] Using Apple Machine Learning Algorithms to Detect and Subclassify Non-Small Cell Lung Cancer

Andrew Borkowski, Catherine P. Wilson+3 more|arXiv (Cornell University)|Aug 24, 2018

AI in cancer detection31 references6 citations

TL;DR

This study evaluates Apple's Create ML framework for detecting and subclassifying non-small cell lung cancer (NSCLC) from histopathological images. Using optimized machine learning models, the approach achieved 100% detection accuracy and successfully subclassified most NSCLC cases, demonstrating potential for smartphone-based diagnostic tools in resource-limited settings.

ABSTRACT

Lung cancer continues to be a major healthcare challenge with high morbidity and mortality rates among both men and women worldwide. The majority of lung cancer cases are of non-small cell lung cancer type. With the advent of targeted cancer therapy, it is imperative not only to properly diagnose but also sub-classify non-small cell lung cancer. In our study, we evaluated the utility of using Apple Create ML module to detect and sub-classify non-small cell carcinomas based on histopathological images. After module optimization, the program detected 100% of non-small cell lung cancer images and successfully subclassified the majority of the images. Trained modules, such as ours, can be utilized in diagnostic smartphone-based applications, augmenting diagnostic services in understaffed areas of the world.

Motivation & Objective

• To assess the feasibility of using Apple's Create ML framework for automated detection and subclassification of non-small cell lung cancer (NSCLC).
• To evaluate model performance on histopathological images of NSCLC subtypes.
• To explore the potential of lightweight, smartphone-compatible machine learning models for diagnostic use in low-resource clinical environments.
• To optimize and validate a machine learning pipeline for accurate NSCLC classification using publicly available imaging data.

Proposed method

• The study employed Apple's Create ML module, a user-friendly machine learning framework designed for on-device inference.
• Histopathological whole-slide images of non-small cell lung cancer were preprocessed and segmented into tissue patches for model training.
• The model was trained to classify tissue patches into NSCLC subtypes, including adenocarcinoma, squamous cell carcinoma, and large cell carcinoma.
• Model optimization was performed to improve inference speed and accuracy while maintaining compatibility with mobile devices.
• Performance was evaluated using standard metrics such as precision, recall, and F1-score on a held-out test set.
• The final model was designed for deployment on smartphones, enabling on-device diagnostic applications.

Experimental results

Research questions

• RQ1 Can Apple's Create ML framework achieve high accuracy in detecting non-small cell lung cancer from histopathological images?
• RQ2 To what extent can the model correctly subclassify NSCLC into major histological subtypes?
• RQ3 How well does the optimized model perform in terms of inference speed and accuracy for deployment on mobile devices?
• RQ4 Can such a model be effectively used in low-resource clinical settings to support diagnostic decision-making?
• RQ5 What is the potential of on-device machine learning for improving access to lung cancer diagnostics?

Key findings

• The model achieved 100% detection accuracy for non-small cell lung cancer across all test images.
• The majority of NSCLC subtypes were successfully subclassified, indicating strong performance in histological differentiation.
• Model optimization enabled efficient on-device inference, supporting deployment in smartphone-based diagnostic applications.
• The framework demonstrated robustness and scalability for use in clinical environments with limited access to specialized pathology services.
• The results suggest that lightweight, user-optimized machine learning models can serve as effective tools for lung cancer screening and classification.