Artificial Intelligence for Lung Cancer Characterization in HIV affected populations in Uganda and Tanzania

点击翻译按钮获取中文摘要

ABSTRACT – Project 3 The age standardized rates (ASRs) show a steady rise in the incidence of lung cancer in Uganda and Tanzania compared to other cancers. Unfortunately, there is no established lung cancer screening program in either of Tanzania or Uganda. The cases of lung cancer recorded have mostly been found incidentally on chest computed tomography (CT) scans done to establish the cause of patients' respiratory symptomatology. This problem of diagnostic specificity is exacerbated in Tanzania and Uganda on account of the high incidence of tuberculosis (TB) which can cause a chronic granulomatous reaction in the lungs manifesting as benign pulmonary nodules on CT and X-rays. Skilled personnel to acquire good quality chest x-ray and CT images and to interpret them is lacking in most tertiary health centers in Uganda and Tanzania. Additionally, the number of people living with HIV AIDS continues to rise, and in 2014, it was reported that Tanzania had 1,411,829 people living with HIV AIDS. However, very little is known about lung cancer and HIV in Africa. With the currently observed increasing incidence rates of lung cancer, there is an urgent need to study the link between lung cancer and HIV in Uganda and Tanzania. An additional intriguing question is whether the same radiographic criteria for lung cancer screening should be uniformly applied across both HIV+ and HIV- patients. Our group has been developing new classes of radiomic (computerized feature analysis of radiographic scans) features for improved discrimination of malignant from benign lung nodules. For instance, we have shown that the tortuosity of nodule vasculature is substantially different between benign and malignant nodules. Additionally, we have shown that radiomic features of the peri-nodular surface (immediately outside the lung nodule on CT and X-rays) were associated with degree of immune response on biopsy tissue specimens. Given that HIV patients tend to have a low immune cell population, a reasonable conjecture is that the radiomic signature on radiographic scans will reflect the absence of an immune signature. In this project we will develop a radiomics based machine classifier called LunIRiS (Lung Image Risk Score) for predicting risk of malignancy for a nodule on a chest CT or X-ray scan. We hypothesize that the new radiomic biomarkers can enable improved non-invasive lung diagnosis in Uganda and Tanzania which has a higher prevalence of TB and hence TB induced granulomas. Additionally, we will seek to employ these tools to identify possibly differences in the radiographic phenotype on CT and chest X-rays between HIV+ and HIV- lung cancer patients and to employ these differences to develop HIV status specific lung cancer screening models. Finally, the fourth objective will be to create a web-based deployment of LunIRiS to enable decision support and teleradiology based services between Cleveland and Uganda and Tanzania for improving lung nodule diagnosis on screening LDCT scans. This partnership will allow for transference of technology and radiology expertise (through the web portal) for improved lung cancer screening in Uganda and Tanzania.