Computational methods for cancer diagnosis and prognosis from FT-IR spectroscopy data

Abstract

Prostate cancer (PCa) is the single most prevalent cancer in US men. PCa diagnosis and prognosis are crucial processes in managing patients and disease. These greatly affect decision-making of treatment and post-treatment strategies. Manual histologic assessment of stained tissue forms gold standard of diagnosis and limits speed and accuracy in clinical practice and research of prostate cancer diagnosis. Outcome prediction by clinical, pathologic, imaging and computational tools outperforms manual ad hoc decisions, but the deficiencies in these tools are apparent and hamper effective and efficient treatment plans. Here, we sought to develop automated tools for cancer pathology using fourier transform infrared (FT-IR) spectroscopy imaging. We employ tissue microarrays (TMAs) to record IR imaging data and adopt elaborated computational and analytical algorithms and statistical methods to recognize patterns and to construct robust systems. We firstly suggest a statistical framework to examine FT-IR imaging from a large population data in appreciation of its design and to identify primary sources of variation. Secondly, we develop cancer detection method by combining FT-IR imaging with microcopy imaging. Extracted morphologic features are excellent in recognizing cancer tissue and robust to staining conditions. Thirdly, a novel decision-support system to aid cancer pathology is introduced and provides an easy access and maintenance of tissues. Lastly, we develop a new prognostic model utilizing FT-IR imaging where stromal chemical features are detected and utilized to characterize cancer progression. The computational and automated methods developed here will prove the utility of FT-IR imaging in cancer pathology and help the development of solid protocols for clinical translation, thereby advancing PCa pathology service today.