The exponential growth and availability of healthcare data coupled with the increasing computational power have enabled the development of AI systems and machine learning algorithms capable of performing advanced clinical care tasks, especially in medical imaging.
These new technologies can learn by automatically extracting features and make reliable predictions without prior instructions. Growing evidence supports the leading role of artificial intelligence (AI) in all cancer imaging pathways, starting from screening to diagnostic and prognostic tasks, dramatically boosting the paradigm of precision medicine.
AI and machine learning approaches have been investigated in oncology in the past, especially in histopathological and molecular diagnosis, and the results in cancer imaging are truly encouraging.
Because of their prominent role in oncologic patients for staging, treatment monitoring, and follow-up, magnetic resonance imaging (MRI) and computed tomography (CT) are the most widely used imaging modalities available today. Moreover, the introduction of advanced imaging techniques such as perfusion CT, MRI, and MRI diffusion-weighted imaging could add functional over morphological data to further characterize tumor phenotype and behavior.
The current AI applications in cancer imaging include optimizing the clinical-radiological workflow (patient screening, image acquisition) and more specific image-based tasks (cancer detection, characterization, and treatment monitoring).
This post will further explore the possible applications of AI in oncology imaging.
1. Clinical-radiological workflow
Due to the ability to analyze a large volume of different data types, such as clinical risk factors, genetic data, and imaging examinations, AI techniques can enable the aggregation of clinical and imaging data to improve screening program efficiency.
As evidenced by recent studies that looked into the impact of the ML model’s clinical practice in identifying individuals at increased risk of breast cancer, breast cancer is undoubtedly a leading area of AI development, particularly in screening practices.
Another study compared the performance of various machine learning (ML)-based techniques in predicting breast cancer risk using clinical and genetic risk factors to the Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) risk prediction model. Decision ML-based models performed better in classifying cancer cases from non-cancer cases, increasing predictive accuracy by 20% to 25%.
Furthermore, significant differences in classification for mammography surveillance were observed between the BOADICEA and risk-based ML models, confirming ML prediction models’ feasibility in the clinical-imaging decision workup. The use of Deep Learning (DL) methods has been shown to significantly impact the reduction of radiation dose in CT scans. They have been used for improving magnetic resonance imaging quality with the potential to decrease acquisition time.
2. Cancer detection
Recent shreds of evidence of AI applications include breast cancer detection in mammography, tomosynthesis, and MRI and the identification of CT lung nodes, brain tumors, and prostate cancer on MRI. Because breast cancer can be masked by healthy breast tissue, breast cancer mammographic detection is a difficult image analysis task.
In a recent study, a DL AI system provided by Google Health company outperformed the radiologists involved in the mammographic screening from multiple centers in the United Kingdom (UK) and United States (US). In the UK test set, the AI system reduced false-positive and false-negative rates by 1.2 percent and 2.7 percent, respectively, and 5.7 percent and 9.4 percent in the US dataset. Moreover, the AI system exceeded six expert radiologists’ average performance who interpreted a sample of 500 randomly selected cases in a controlled study.
In lung cancer, recent research showed that a DL automatic detection algorithm achieved higher performance than the radiologist group in detecting malignant pulmonary nodules on chest radiographs. Moreover, radiologists’ performance improved when the DL algorithm was used as a second reader.
3. Tumor segmentation, characterization, and staging
Segmentation represents one of the most challenging oncological image analysis tasks, and AI algorithms have allowed the development of systems that can enable automatic tumor segmentation. Recently, deep learning networks (DL networks), such as CNN, have been used in segmentation tasks to improve radiotherapy treatment planning, volume measurements, and disease progression monitoring.
Indeed, increasing evidence has highlighted DL models’ high performance in performing fully automated whole-breast segmentation to obtain reliable and robust quantitative imaging analysis methods. High-performance AI algorithms for multi-dimensional data have enabled the extraction and analysis of radiomic biomarkers reflecting image tumor heterogeneity, allowing for more precise cancer imaging diagnosis and staging.
Staging is an important part of the oncological workflow because it allows doctors to pinpoint the best treatment options in a personalized and precise manner. Recent pilot studies on the staging of primary tumor size, lymph node involvement, and distant metastasis have been conducted in this regard.
For example, a group investigated the clinical feasibility of combining radionics and machine learning on MR images to detect deep myometrial invasion in endometrial cancer in a clinical setting; integrating the developed ML algorithm improved radiologist accuracy from 82 percent to 100 percent.
4. Treatment monitoring
Temporal tumor follow-up and treatment response are active AI research fields, intending to develop accurate models for evaluating effective anticancer therapies that improve patients’ progression-free survival.
In this area, excellent results have been obtained using AI radionics MRI-based models in predicting survival and recurrence-free survival in breast cancer. Furthermore, AI models have been tested in breast cancer imaging to assess predictive image-based phenotypes for precision medicine, specifically to predict neoadjuvant chemotherapy response (NAC).
A recent study explored the usefulness of a combined radionics MRI-based and molecular subtype-based ML model in assessing the complete pathological response (pCR) to NAC. The AI model accurately predicted pCR on MRI with an AUC of 0.88 and showed that predicting pCR increased when radionics features were combined with molecular subtype compared to the solely molecular subtype results.
Despite the successes of AI within cancer imaging, several limitations and hurdles must be overcome before widespread clinical adoption. With the increasing demand for CT and MR imaging, care providers are constantly generating large amounts of data. However, such data are rarely curated in terms of labeling, annotations, segmentations, quality assurance, or fitness for the problem at hand. The curation of medical data represents a major obstacle in developing automated clinical solutions, because it requires trained professionals, making the process expensive in both time and cost. Another limitation includes the interpretability of AI and the ability to interrogate such methods for reasons behind a specific outcome, as well as the anticipation of failures.
