The Coding of Roentgen Images for Computer Analysis as Applied to Lung Cancer

This paper will describe a concept of converting the visual images on roentgenograms into numerical sequences that can be manipulated and evaluated by the digital computer and will report the results of employing this system to determine the significance of certain radiographic findings in lung cancer. The development of such a coding system for conveying radiographic information into an electronic data-processing system makes possible exploration of the use of the digital computer as an aid in radiologic diagnosis. The built-in capacity to retain vast numbers of facts, to accept instruction from the physician to compare new facts with its stored information, to report the result of such comparison in the form of statistical probability, and to carry out these functions with great speed and accuracy makes the usefulness of the computer most obvious. The system of communication reported here, which permits high-speed systematic evaluation of radiological data, is a logical approach to the control of a segment of exponentially expanding medical knowledge. We have chosen to apply this concept to roentgenograms of lung cancer because, against a background of air density, the intimate details of the relationship between tumor and host may be faithfully reproduced roentgenographically. Parenthetically, it may be stated that similar density ranges exist in the relationships between bone and soft tissue and that an equally effective descriptive system has been evolved for bone sarcomas (1, 2). For this study, the roentgenograms of 541 cases of primary lung cancer were made available from the files of the Department of Radiology of the State University of Iowa. The patients who were radiographed were seen during the years 1939 to 1952, inclusive. In all, the clinical diagnosis was primary lung cancer. In 66.2 per cent this diagnosis was confirmed by histologic examination of tissue; 43.4 per cent of the patients were treated by operative exploration, and 19.4 per cent underwent either resection of a segment of lung or a pneumonectomy. The five-year survival data for this group of cases are shown in Table I. Less than 1 per cent of the total number were lost to follow-up. The absolute survival rate of 1.3 per cent for this highly malignant tumor is even lower than that of several other large series of reported cases (Table II), reflecting the advanced stage of the disease in many of the patients referred to this state-wide diagnostic and treatment center (3, 4, 5). This advanced stage of disease is also reflected in surgical resectability (Graph I), where for the four studies analyzed, at least, the relationship between rate of resectability and rate of five-year survival is exponential (for each 10 per cent increase in resectability, survival rate increases by a factor of approximately 2.5×).