Using Naïve Bayesian Analysis to Determine Imaging Characteristics of KRAS Mutations in Metastatic Colon Cancer

Area of Science:

• Oncology
• Radiology
• Computational Biology

Background:

• KRAS mutation status significantly impacts colon cancer prognosis and treatment strategies.
• Liver metastasis in colon cancer is influenced by patient genotype, particularly Ras status.

Purpose of the Study:

• To apply word frequency analysis and a naive Bayes classifier on radiology reports.
• To extract distinguishing imaging descriptors for wild-type versus KRAS-mutated colon cancer patients.
• To correlate specific radiological terms with KRAS mutation status.

Main Methods:

• Analysis of over 32,000 radiology reports from 299 colon adenocarcinoma patients.
• SNaPshot mutation analysis to categorize patients into wild-type (147) or KRAS mutation (152) groups.
• Development of a naive Bayes classifier to determine word probability within each group.

Main Results:

• Words like "several", "innumerable", "confluent", and "numerous" were more frequent in KRAS mutation reports (p < 0.01).
• Words such as "few", "discrete", and "[no] recurrent" were more frequent in wild-type reports (p = 0.03).
• Differing word frequencies in reports correlate with disease course, tumor burden, and therapy implications.

Conclusions:

• Radiology report language varies significantly between KRAS-mutated and wild-type colon adenocarcinoma.
• Probabilistic word analysis of radiology reports can identify unique tumor characteristics and disease course.
• This approach has potential applications in radiology, pathology, and clinical notes for improved patient management.